ProtNLM Prediction Evaluation

Extracellular matrix is correct for full-length OLFML2A: mouse photomedin-1 (Q8BHP7, same PANTHER subfamily PTHR23192:SF29) is experimentally secreted and binds chondroitin sulphate-E and heparin (PMID:15836428), and the review of the full-length red colobus product A0A8C9H9A8 proposes this location. This accession, however, is the short isoform that starts at an alternative first exon and lacks the region carrying the signal peptide, both CXCXCX9C motifs and the coiled coil. Without a signal peptide it is not expected to enter the classical secretory pathway, and there is no evidence of release by another route. The gene review marks the parent location, extracellular region (GO:0005576, an ancestor of GO:0031012 by OLS4), as over-annotated on this product for that reason, which equally refutes the more specific matrix location. The prediction transfers a full-length-protein property to a product lacking the required targeting segment. The term is absent from GOA for both products of the gene, so it would have been novel.

This accession is the 439-residue short isoform that starts at an alternative first exon (counterpart of human isoform 3). Its residues 9-439 align to residues 225-655 of the full-length product of the same gene, A0A8C9H9A8, which carries a SignalP signal peptide at 1-27; A0A8C9H4D2 has no counterpart of that region and so lacks the signal peptide, both CXCXCX9C motifs and the coiled coil, and is not expected to enter the classical secretory pathway. That absence does not settle the process term, because no mechanism of OLFML2A participation in extracellular matrix organization has been demonstrated, so it cannot be shown that the missing segment accounts for that participation, and no experiment refutes a contribution by this product. The only experimental basis is an MGI IDA annotation of mouse Olfml2a (Q8BHP7, PTHR23192:SF29) with the acts_upstream_of_or_within qualifier, from a systematic screen of candidate secreted proteins for matrix-assembling activities (PMID:18757743), which the mouse review leaves UNDECIDED because the Olfml2a-specific result is not visible. Secretion and glycosaminoglycan binding of full-length photomedin-1 (PMID:15836428) make a matrix role plausible for the full-length product, but no assembly result exists for any ortholog. Whether this short product is translated or released has not been tested. The prediction can be neither validated nor refuted. Neither product of the gene carries the term in GOA.

GOA has no annotations for this protein, so an ER term would be novel, but the available evidence points to the nucleus instead. A0A8B8L1Z3 is in PTHR45496:SF19 and is closest among Arabidopsis proteins to SDJ1/DNAJ1/EIP9 (about 46% identity over a similar length). PAINT asserts nucleus at node PTN002896004, which spans the SDJ1, SDJ2, SDJ3 and ADMETOS subfamilies. EIP9/SDJ1 heterodimers with EMF1 localize to the nucleus, and SDJ1/DNAJ1 is recruited to methylated DNA by the SUVH1/SUVH3 methyl readers, where it enhances transcription. The UniProt record shows no signal peptide or transmembrane segment, only an N-terminal J domain followed by disordered regions. The J domain lacks the HPD motif, which argues against the protein acting as an Hsp70 (for example BiP) cochaperone at the ER. The ER label most likely reflects the many ER-resident J-domain proteins rather than any feature of this subfamily. No literature reports ER localization for any SF19 member.

The target has a complete RasGAP domain (IPR001936) with the RasGAP conserved site and keeps the catalytic arginine finger (R483 in a TFRGNS loop identical to raskol and matching neurofibromin). Mutational work on neurofibromin and p120GAP shows that the finger loop stabilises the GTP-hydrolysis transition state (PMID:12787671). PAINT placed GTPase activator activity as an IBD at the PTHR10194 root node PTN008939999, from which this protein descends, and within subfamily SF60 purified yeast Bud2p stimulates GTP hydrolysis by its Ras-like target (PMID:8371782). This is a phylogenetic and mechanistic inference; the mosquito protein has not been assayed, and the general term is appropriate because the substrate GTPase is unknown. The re-reviewed gene review proposes the same term as a NEW core function. QuickGO returns no annotation for A0A6I8TLE4 or the other AAEL019982 products, and the UniProt keyword-derived GO line is not in GOA, so the prediction is novel.

The target is a 156-residue fragment of bullfrog TLL1: residues 30-156 match the last 127 residues (CUB5) of the complete bullfrog TLL1 model at 126/127 identity, and identity to other tolloids follows the TLL1 orthogroup. The fragment carries only a CUB domain, with no S1 serine-protease domain and no His-Asp-Ser triad. That excludes the ovochymase-like egg-envelope protease mechanism the prediction implies, for both the fragment and the gene. The complete TLL1 product is an astacin-type metalloprotease that processes procollagen C-propeptides and cleaves chordin, and the PTHR24251 family review finds that single fertilization is not a family-wide process. These facts exclude the predicted mechanism, not every way TLL1 could take part in fertilization. TLL1 has no known mechanism of participation in fertilization, so no missing feature can be shown to be required, and a non-catalytic CUB-mediated binding or extracellular-matrix role at the egg surface is neither shown nor excluded. PubMed searches for tolloid-like proteins in fertilization and for bullfrog tolloids returned no reports either way. The evidence is an absence of support, not a positive refutation, so the prediction is uncertain.

The sequence ProtNLM2 was given is a 156-residue fragment of bullfrog TLL1. Residues 30-156 are the C-terminal CUB5 domain, at 126/127 identity to the last 127 residues of the complete 1005-aa bullfrog TLL1 model, and residues 1-29 align to no tolloid model. The fragment lacks the TLL1 N-terminal signal peptide, the feature that routes the complete product into the secretory pathway and so to the extracellular region. Its UniProt record has no signal-peptide or transmembrane feature, only the CUB domain at 31-147. Extracellular region is true of the complete TLL1 product (human TLL1 activity is measured in the extracellular medium of transfected cells, and PANTHER places the term at the tolloid node), but the product encoded by this accession has no secretion signal, so the location is not supported for this sequence. GOA has no annotation for this accession.

Catalytic activity is a novel term for this gene, whose only molecular-function annotation (on the identical entry F4JTN4) is ND. Local alignment maps the curated phosphohistidine nucleophile, proton donor/acceptor and transition-state stabilizer of human PGAM1 and E. coli GpmA onto His17, Glu109 and His173 of Q0WW53, so the catalytic apparatus of the histidine-phosphatase fold is retained and enzymatic activity is plausible. However, these comparators lie outside PANTHER PTHR47821, no member of that family has been assayed, and PubMed searches for At4g38370, Q0WW53 and F4JTN4 return no functional study. Residue retention does not prove activity, and both gene reviews decline to assert any molecular-function term, so the prediction can be neither validated nor refuted.

No GOA annotation on F4JLB7 covers phosphorylation, so the prediction would be novel. The protein lacks a protein kinase domain (OpenScientist domain and motif analysis), so it cannot phosphorylate a substrate itself. A noncatalytic LRR receptor-like protein could still take part in phosphorylation-based signaling with a partner receptor kinase, as its sister-clade paralog TMM does with ERECTA-family kinases, but that relative is in a different subfamily and its specific functions are not transferable. The gene review accepts only broad signaling receptor activity and signal transduction, with no known ligand, partner or pathway. A PubMed search for At4g28560 returns no experimental papers. Nothing on the target establishes or excludes a role in phosphorylation, so the prediction remains uncertain.

Kinase activity is refuted by the protein's architecture. The 450-residue sequence carries only LRR signatures (InterPro IPR001611, IPR032675; Pfam LRR_1, LRR_8), a signal peptide and a C-terminal hydrophobic anchor with essentially no cytoplasmic tail, leaving no room for a cytoplasmic kinase domain. The OpenScientist analysis finds no protein-kinase Pfam or InterPro domain and no ordered catalytic motifs; the isolated DFG-like tripeptide lies within the LRR solenoid. This is a computational assessment, not a biochemical assay. The re-reviewed gene review treats F4JLB7 as a noncatalytic cell-surface receptor-like protein, and neither GOA nor the PAINT node for PTHR48004 assigns any kinase activity. The prediction is NPI.

The 661-residue target has the complete EF4/LepA architecture and is placed in the chloroplastic LepA subfamily PTHR43512:SF4, sharing 82.9% identity with Arabidopsis cpLEPA (At5g08650) and clearly less with the mitochondrial GUF1 subfamily SF7. Arabidopsis cpLEPA was shown by fractionation of intact chloroplasts to reside mainly on thylakoids with a minor stromal pool, and its loss impairs chloroplast protein synthesis (PMID:23166764), which grounds transfer of chloroplast localization within the plant SF4 branch. The re-reviewed gene review agrees, replacing the HAMAP-derived mitochondrial inner membrane and matrix rows with chloroplast thylakoid membrane and chloroplast stroma; those mitochondrial rows are the only cellular-component annotations in GOA, so chloroplast is novel for this gene. The term is correct though less specific than the suborganellar locations the evidence supports, and it rests on subfamily transfer rather than a targeting assay on the Artemisia protein.

Scored for the 134-residue Q2U1U6 translation, which is a truncated gene model: it corresponds only to residues 280-408 of the ~408-aa full-length product (A. flavus ortholog A0A7U2MCM8, 128/129 identity), because the AO090138000091 model starts at an ATG inside a 65-nt intron and omits the ~279-codon first exon. In the full-length product the chondroitin AC/alginate lyase alpha/alpha-toroid domain (IPR008929, SSF48230, Gene3D 1.50.10.100) spans residues ~62-371, so this accession carries only roughly the last third of the toroid and is unlikely to be a competent lyase. That does not refute the process term: no participation mechanism in polysaccharide catabolism is known for this gene even at full length (the lyase activity is a fold-level, unverified inference with unknown substrate, no Pfam/PANTHER/CAZy family, no GOA annotations, no literature), so the process cannot be shown to depend solely on the missing part of the fold. A non-catalytic structural or binding contribution by the fragment is neither shown nor excluded. With no positive evidence for or against participation, the prediction is uncertain.

The protein's only signature is the chondroitin AC/alginate lyase alpha/alpha-toroid fold (IPR008929, SSF48230), whose characterized members cleave uronic-acid-containing polysaccharides by beta-elimination rather than hydrolysis. The full-length product of the truncated gene model has no Pfam, PANTHER or CAZy glycoside hydrolase family assignment, and the re-reviewed gene's core function is a fold-level carbon-oxygen lyase activity acting on polysaccharides. No experimental evidence or literature exists for any glycosidase activity. The O-glycosyl hydrolase prediction transfers a hydrolytic activity onto a lyase fold and is discordant with the only available domain evidence.

The target carries the Xin actin-binding repeat (IPR012510) and the XIRP2 subfamily assignment (PTHR22591:SF1). Xin repeats, including those of human XIRP2, bind actin filaments directly and arrange them into networks (PMID:15454575), which supports actin cytoskeleton organization by phylogenetic transfer. The exact term is already in GOA for this gene (InterPro2GO), and the gene review refines it to actin filament organization (GO:0007015), which the gene also carries by IBA. The prediction is correct but not novel.

Mouse Xirp2 (mXinbeta) localizes to the intercalated disc of the heart (PMID:20360251), a cell-cell junction, and the bovine target is its XIRP2-subfamily ortholog. Cell junction is therefore correct by orthology. The exact term is already in GOA for this gene (InterPro2GO), and the gene review refines it to cell-cell junction (GO:0005911), which the gene also carries by Ensembl Compara transfer. The prediction is correct but not novel.

Xin repeats are experimentally established actin-binding modules, and human XIRP2 repeats share this activity (PMID:15454575). The bovine target contains these repeats (PROSITE PRU00721). Actin binding is already in GOA for this gene (combined automated annotation, GO_REF:0000120), and the gene review refines it to actin filament binding (GO:0051015), which the gene also carries by IBA. The prediction is correct but not novel.

No evidence places C. elegans mcm-4, or this 74-residue C-terminal winged-helix fragment of it (canonical MCM-4 residues 750-823), in Pol II transcription initiation. The characterized mcm-4 functions are DNA replication and the replication checkpoint (PMID:21146520), and neither the gene's existing annotations nor its re-reviewed gene record contain a transcription term, so the prediction is novel but unsupported. It is not refuted outright: vertebrate MCM subunits copurify with the Pol II holoenzyme (PMID:10454562), but that association was mapped to an N-terminal segment of MCM2 (residues 168-230; PMID:12392551), it is complex co-purification rather than evidence that MCM-4 performs a step of initiation, and nothing ties it to the MCM4 C-terminal winged-helix region that makes up this product. Whether the fragment is even translated is unknown. The OpenScientist analysis judged the prediction NPI from the replicative-helicase identity of MCM-4; its targeted moonlighting search found no primary evidence for an MCM role in Pol II transcription and did not retrieve the holoenzyme co-purification reports (PMID:10454562, PMID:12392551), which leave the term unsupported rather than refuted.

Nucleus is not novel at the gene level: the canonical mcm-4 product Q95XQ8 carries GO:0005634 by IDA from PMID:21146520, where MCM-4::mCherry was nuclear and associated with chromatin from late anaphase, and GOA keeps the gene's annotations on that representative product. Correctness for this accession is unresolved. A0A061AL94 is only the C-terminal winged-helix region (canonical residues 750-823) from an internal-start gene model, its translation in vivo has not been shown, and the gene review deliberately does not transfer full-length MCM-4 localization or replication functions to it. The prediction therefore agrees with the gene's annotation but lacks support for the deposited sequence itself.

The target is the elephant shark ESRP1 ortholog: it carries the ESRP family and ESRP1 RRM1 InterPro signatures and groups with tetrapod ESRP1 in a neighbour-joining tree. Its sequence contains the peptide GLSPPPCKLPCLSPP (UniProt sequence), which experiments on mouse Esrp1 showed is sufficient for nuclear localization, while isoforms lacking CKLP are mainly cytoplasmic (PMID:28634384). Nucleus is therefore a justified ortholog transfer, though not exclusive of a cytoplasmic isoform pool. The gene already carries nucleus in GOA (accepted in the gene review), so the prediction is correct but not novel.

The target is placed in the CNNM4 subfamily (PANTHER PTHR12064:SF26) and has a signal peptide, four predicted transmembrane helices within the CNNM transmembrane domain, and a cytoplasmic CBS domain, the architecture of a multi-pass plasma-membrane transporter. Mouse CNNM4 localizes to the basolateral membrane of intestinal epithelia and ameloblasts (PMID:24339795), and endogenous CNNM4 in Madin-Darby canine kidney cells is basolateral (PMID:25449265), which is direct evidence in the dog. Plasma membrane is already annotated to the target, and the more specific basolateral plasma membrane (GO:0016323, is_a descendant of GO:0005886 per OLS4) is an existing annotation that the gene review accepts. The prediction is correct but less specific than that supported annotation.

The target belongs to the CNNM4 subfamily (PTHR12064:SF26) and carries the CNNM transmembrane and CBS domains. Mouse CNNM4 extrudes Mg2+ from cells in a Na+-dependent manner, mediating transcellular Mg2+ transport (PMID:24339795), and structural work on human CNNM4 describes Mg2+/Na+ exchange (PMID:42497866); the dog protein is in the same subfamily, so the activity transfers by phylogenetic inference rather than direct observation. Generic transmembrane transporter activity is true but uninformative: the target already carries magnesium ion transmembrane transporter activity (GO:0015095, a descendant of GO:0022857 per OLS4) and sodium ion transmembrane transporter activity, and the gene review refines these to magnesium:sodium antiporter activity. The prediction is therefore less precise than existing supported annotations.

CNNM4 mediates basolateral Mg2+ extrusion coupled to Na+ in mouse epithelia (PMID:24339795), and the dog target inherits this function through its CNNM4 subfamily placement (PTHR12064:SF26) and conserved transmembrane and CBS domain architecture. Monoatomic ion transport is not itself annotated, but the target already carries magnesium ion transport (GO:0015693), magnesium ion transmembrane transport and sodium ion transmembrane transport, which the gene review accepts; OLS4 confirms GO:0015693 is a descendant of GO:0006811 via metal ion and monoatomic cation transport. The prediction is correct but discards the established ion specificity, so it is less precise than existing annotations.

This accession, the exact sequence ProtNLM2 was given, cannot have tRNA pseudouridine synthase activity because it lacks the catalytic aspartate that every TruB-family pseudouridine synthase activity requires. The 142-residue gene model aligns to human TRUB2 residues 178-307 and has no counterpart of human residues 1-177, which contain the active-site nucleophile D98 (equivalent to E. coli TruB D48) and most of the Pseudouridine synthase II N-terminal catalytic domain; only residues 1-54 of that domain remain (UniProt DOMAIN 1..54). For the gene, GO:0106029 would be a novel refinement of the pseudouridine synthase activity (GO:0009982) it already carries, and even at gene level a tRNA-specific activity for the TRUB2 lineage is unsettled: recombinant human TRUB2 has tRNA Psi55 activity (PMID:33023933), but a cell-wide tRNA pseudouridine map found no noticeable TRUB2 activity (PMID:41136621), in-cell evidence centres on mitochondrial mRNA sites (PMID:27974379), and the bilaterian TRUB2 PAINT node asserts only the general activity.

The quail protein is in PANTHER subfamily PTHR10680:SF14 with experimentally characterized mammalian PAM and carries a complete NHL-repeat PAL domain. The crystal structure and mutagenesis of the rat PAL catalytic core establish N-dealkylation of the peptidyl-alpha-hydroxyglycine to the amidated peptide and glyoxylate (PMID:19604476), which justifies transfer of this activity within the subfamily. The gene already carries this exact term (GO:0004598, from EC 4.3.2.5 mapping), and its gene review accepts it as a core function, so the prediction is correct but not novel.

Ascorbate is the reductant co-substrate of the PHM module, which carries the copper type II ascorbate-dependent monooxygenase domains (IPR000323, IPR014784). Stopped-flow reduction kinetics on rat PHM, from the same PANTHER subfamily (SF14), show saturable ascorbate dependence (KD about 1.1 mM), indicating that ascorbate binds the enzyme to reduce the copper centres (PMID:35380039). The same study notes that crystallography has not identified a discrete ascorbate site, so the binding is kinetically defined and of low affinity. The gene's annotations include the ascorbate-dependent monooxygenase activities (GO:0016715, GO:0004504) but no ascorbate binding term, and the PTHR10680 family review assigns L-ascorbic acid binding to PHM-bearing members. The prediction is therefore a supported, novel molecular function for this protein, though one that adds little beyond the co-substrate role implicit in its catalytic activity.

The target has the PTEN-like, C2, and J-domain architecture of auxilin/DNAJC6, but the PTEN-like region is a conserved pseudophosphatase rather than an active phosphatase. Structural work on bovine auxilin shows that this region lacks phosphatase activity and instead participates in membrane binding (PMID:20826345), and the focused sequence comparison found the dead C-X3-R P-loop across human, mouse, and zebrafish DNAJC6 instead of PTEN's active C-X5-R loop. The predicted dephosphorylation biological-process term is an over-propagation from a catalytic PTEN-like ancestor: no evidence establishes that dnajc6 itself catalyzes or directly executes a dephosphorylation step.

DCXR is an oxidoreductase, but the generic term is an ancestor of the more specific activities the gene already carries in GOA and that the gene review accepts as core, L-xylulose reductase (NADPH) activity (GO:0050038) and carbonyl reductase (NADPH) activity (GO:0004090), both by IBA from PAINT node PTN001211870. Those activities are measured on purified mammalian DCXR and on the nematode SF3 ortholog DHS-21, and Q567K5 aligns gaplessly to human DCXR with the catalytic triad conserved. No zebrafish enzyme assay was found in a PubMed search. The correct broad class does not validate the separate FabG function paragraph.

Direct assays establish sequence-specific promoter binding by the Her7:Hes6 heterodimer. The predicted term is already in GOA (InterPro IEA) and is an ancestor (verified in OLS) of RNA polymerase II cis-regulatory region sequence-specific DNA binding (GO:0000978), which the gene carries by IBA and the gene review accepts. Generic DNA binding is valid for the complex contribution but must not be interpreted as autonomous Hes6 homodimer binding.

Hes6-dependent negative transcriptional regulation is established by reporter, genetic and biochemical experiments. The predicted term is already in GOA (IEA) and is an ancestor (verified in OLS) of negative regulation of transcription by RNA polymerase II (GO:0000122), which the gene carries by IBA and the gene review accepts as core, so the prediction is correct but less specific.

Both homo- and heterodimerization are directly demonstrated for zebrafish Hes6, which acts as the hub of the Her/Hes dimer network. The predicted term is already present in GOA (InterPro IEA) and is the parent (verified in OLS) of protein heterodimerization activity (GO:0046982), which the gene carries by IPI and which the gene review treats as its core molecular activity, and of protein homodimerization activity. The prediction is therefore correct but materially less specific than existing supported curation.

Nucleic acid binding is correct: Q9RSY6 has five S1/OB-fold single-stranded nucleic acid binding domains, sits in the same PANTHER subfamily (PTHR10724:SF7) as E. coli RpsA, whose S1 binds RNA (PMID:9451450), and the purified D. radiodurans protein binds RNA oligonucleotides in vitro (PMID:35856907). The gene already carries the more specific GO:0003729 mRNA binding (IBA, accepted as core in the gene review), a descendant of GO:0003676 (OLS4 ancestry); the gene review refines the identical InterPro2GO nucleic acid binding row to mRNA binding. The prediction is supported but less precise than the existing annotation.

Ribosome is correct: Q9RSY6 belongs to the bacterial bS1 family (PANTHER PTHR10724:SF7, with E. coli RpsA) and bS1 is a component of the 30S subunit, binding it via its N-terminal S1 domains. The gene already carries GO:0022627 cytosolic small ribosomal subunit (IBA, accepted in the gene review), and GO:0005840 ribosome is an ancestor of that term (OLS4 ancestry). The prediction is supported but less precise than the existing annotation.

Q3KN55 is the native short Ank2-PE product of the Drosophila neuronal ankyrin locus, not a latrotoxin. The prediction metadata identify alpha-latrocrustotoxin Q9XZC0 as the phmmer donor (model score 0.49; phmmer score 293.2). The experimentally determined latrotoxin structure requires a central membrane-insertion domain in addition to ankyrin-like repeats; generic repeat similarity does not transfer its evolved inter-organism toxic function to the fly cellular ankyrin. The target record instead shows a low-complexity N terminus and ankyrin repeats, with no evidence for the toxin-specific architecture or venom role. This is a biologically incompatible functional transfer, not a demonstrated paralog relationship or a FlyBase annotation error. No equivalent toxin term occurs in the target GOA.

Ank2-dependent synaptic stability and neurotransmission make a role in vesicle release conceivable, but they do not establish exocytosis by the exact 697-residue Ank2-PE isoform. The principal experiments characterize giant Ank2 isoforms and their extended C-terminal microtubule-organizing region. The API metadata indicate human ankyrin-1 P16157 as the donor for this claim (model score 0.53; phmmer score 889.4), rather than the latrotoxin donor used for the toxin claim. Donor exocytosis annotation and synaptic phenotypes are insufficient to resolve transfer across these distinct proteins and isoforms. An Ank2-PE-specific secretion assay or supported interaction with the vesicle-fusion apparatus is missing, and a PubMed search of Drosophila Ank2 found no exocytosis or secretion study, so neither correctness nor biological refutation is justified. The term is absent from the GOA of both Q3KN55 and the sibling Ank2 product M9MRX4.

Residence in the extracellular region requires a route out of the cell, normally an N-terminal signal peptide, and the 697-residue Ank2-PE product scored here has none. Its N terminus begins MAQFVTHIQPTLMEASQGHVPHHH and continues as a histidine-rich, disordered, low-complexity segment (MobiDB-lite disordered region 17-112) with no hydrophobic core, and the record carries no signal peptide or transmembrane feature. Q3KN55 consists of that unique N terminus, the first 13 ankyrin repeats shared with the giant isoforms, and a short tail, none of which supplies a secretion signal. The model's donor is the secreted spider toxin Q9XZC0, whose compartment cannot be transferred through shared ankyrin repeats. Characterized Ank2 isoforms are intracellular membrane-cytoskeleton adaptors, and the Ank2 ankyrin-repeat domain that Ank2-PE retains is described as part of an intracellular linker to transmembrane partners. Neither Q3KN55 nor the sibling Ank2 product M9MRX4 carries an extracellular term in GOA.

Phosphorylation is plausible for this EcKL/CHK-kinase-like protein, but the available family study proposes rather than demonstrates the relevant reactions across this diverse family. No CG31099 substrate, product or catalytic assay is established in the accessible evidence, and the existing ecdysteroid 22-kinase TAS annotation is itself unresolved. Domain nomenclature and agreement with that annotation therefore cannot independently validate the process, although there is no evidence refuting it. CG31099 lies in PANTHER subfamily PTHR11012:SF6 with only the uncharacterized CG6830, outside the subfamilies of both characterized ecdysteroid 22-kinases, so no phylogenetic transfer supports it. The term would be novel relative to GOA, which carries no process annotation for this gene and whose ecdysteroid 22-kinase activity term does not entail phosphorylation by is_a or part_of.

Lipid transport is not established for CG31606. The raw structural donor is mammalian apolipoprotein A-V, but the strongly asymmetric TM scores do not establish target orthology or a conserved lipid-carrying mechanism. A signal peptide supports secretion rather than cargo specificity. Target lipid transport or convincing subfamily evidence is missing, and the claim is neither validated nor refuted.

Extracellular localization is a reasonable inference from the cleavable N-terminal signal peptide and soluble mature sequence lacking a membrane anchor or canonical ER-retention terminus. It is not an experimentally measured compartment, and the bacterial IpgD donor is not its biological justification. This broad localization is absent from the ND-only GOA record, so it is a supported novel sequence-based prediction. It does not establish extracellular-matrix residence or lipid handling.

No lipid-binding assay or defensible apolipoprotein subfamily placement is available for CG31606. The partial structural match to seal apolipoprotein A-V is a hypothesis lead rather than conserved ligand-binding evidence. Secretion does not identify a ligand. The exact binding claim remains unresolved.

No evidence connects CG31606 to lipoprotein metabolism. The structurally matched apolipoprotein donor is not established as an ortholog, and a predicted signal peptide does not identify a pathway. Target perturbation, biochemical context or sound comparative functional transfer is needed to evaluate this process.

X2JE45 is the native 799-residue PC isoform. It retains the catalytic residues corresponding to human GBA2 Glu527 and Asp677, but replaces PB residues 1–177 with a distinct 28-residue N-terminus that alters the GH116 N-terminal region. Human GBA2 isoform experiments establish that such noncatalytic-domain changes can impair activity; they do not demonstrate that this different fly isoform is inactive. The specific missing evidence is activity or fold integrity of native PC, rather than uncertainty about the GBA2 family as a whole. The term is not novel, because X2JE45 already carries it by InterPro2GO IEA and the gene's PB product (Q7KT91) carries the more specific beta-glucosidase activity (GO:0008422) by IBA; the gene review leaves the X2JE45 row UNDECIDED for the same isoform reason. No experimental study of CG33090 or of any Drosophila GBA2 product was found in PubMed.

The phosphotransferase prediction agrees with CG33116 membership in the choline/ethanolamine phosphotransferase subfamily established in a primary comparative study. The PAINT assignment of ethanolaminephosphotransferase activity is more specific and is a reasonable conserved-reaction inference for this ER protein. The exact substrate activity has not been directly measured in the cited localization study. The predicted term is already in GOA as an InterPro-derived IEA, and the gene review refines it to GO:0004307, which the gene carries by IBA.

Membrane residence is supported by the membrane phosphotransferase architecture and direct localization of CG33116/dCCS1 to the ER. The existing ER-membrane annotation specifies the relevant compartment and is more informative than membrane. The predicted term is already in GOA by IEA, and the gene review refines it to GO:0005789.

The predicted phospholipid biosynthetic process is consistent with the conserved choline/ethanolamine phosphotransferase subfamily and ER localization. The curated phosphatidylethanolamine biosynthetic process is the supported, more specific pathway assignment. This inference concerns glycerophospholipid synthesis, distinct from the ceramide phosphoethanolamine synthesis experimentally assigned to CG4585. The predicted term is already in GOA by IEA, and the gene review refines it to GO:0006646, which the gene carries by IBA.

The source reports structural matches to macaque GM2 activator Q8HXX6, with TM-align scores approximately 0.59/0.55. Current target annotations identify DUF1091, not an experimentally established GM2-activator mechanism. A structurally similar secreted fold could bind lipids or proteins, but the exact claim requires evidence of direct stimulation of a defined enzyme. No target enzyme, activation assay or justified conserved interaction has been demonstrated for CG33453; secretion and a donor annotation alone are insufficient. No equivalent positive target annotation exists.

The structural donor for this prediction is human GM2 activator P17900. That protein presents ganglioside GM2 for lysosomal hydrolysis, a substrate-specific pathway that cannot be transferred solely from a moderate fold match. Primary lipid-analysis and transgenic-reconstitution work shows that gangliosides are not intrinsically detectable in ordinary Drosophila and must be induced using introduced synthesis enzymes and supplied precursor. This contradicts assignment of an endogenous ganglioside-catabolism role to native CG33453. It does not claim that flies cannot sialylate glycoproteins or that an engineered/exogenous substrate could never be degraded. No equivalent positive target annotation exists.

X2JEK1 is the 176-residue CG34171-PC isoform, not the 292-residue PB product. Two alignment modes and two active S1-domain references place H72 and D124 at conserved catalytic positions, but the entire catalytic-serine region is absent from the short product. The longer product has F222 at that position; that is a separate substitution and must not be assigned to the missing short-product sequence. This contradicts the predicted intrinsic activity and agrees with the explicit NOT IKR annotation to the same term. The raw prediction records Q66TN7 as its donor at model score 0.99, but its hit coordinates and historical input sequence are unavailable. The decisive evidence is the reproduced catalytic-region defect, not agreement with a database label or the SPH synonym.

Nuclear localization is a reasonable conserved inference from the diagnostic Swd2/WDR82 family assignment and curated chromatin/COMPASS placement. Primary WDR82 experiments establish chromatin association and Set1-complex membership, providing the biological basis for that transfer. CG3515 is a separate fly paralog from Wdr82/CG17293, so these are not claimed as direct target experiments. The nucleus term is already present, making this supported prediction non-novel.

The exact target is the single 245-residue CG43124-PA polypeptide. Reproducible PF00089 profile alignment maps the essential trypsin catalytic histidine and aspartate to target Ser69 and Asn112 within a significant homologous segment; active bovine trypsin provides a successful positive control. The catalytic-serine region is poorly aligned, so a third substitution is not asserted. These two replacements contradict conventional serine endopeptidase chemistry and support a noncatalytic homolog. The original prediction has model score 0.98 and phmmer donor Q90WD8, toad ovochymase-2; fold similarity to this protease does not restore the missing target catalytic residues. Target GOA contains an explicit NOT annotation for the same activity, consistent with this independent analysis. Opposite polarity is not an equivalent existing positive annotation or CNN. Pseudoenzyme overannotation is supported; prediction-time sequence identity and the native noncatalytic function remain unestablished.

The exact 910-residue Q8IP30 sequence has a protease-like domain with glycine replacing the catalytic serine, supported by substantial-coverage alignments to two active trypsin controls. The controls agree on the additional His-to-Glu and Asp-to-Asn site mappings, although the histidine region is more divergent. These residue observations, rather than the SPH name or an automated domain caution alone, refute conventional serine endopeptidase catalysis. Four existing GOA NOT annotations are concordant negative annotations, not evidence that the positive prediction is correct or non-novel. The source gives model score 0.99 and phmmer donor Q7Z410 with score 226.0; donor similarity cannot restore the missing nucleophile. Gene-specific RNAi establishes cellular immune regulation, without assaying purified-protein proteolysis or resolving a direct mechanism.

Nucleic acid binding is supported by the R3H domain and conserved RNA-helicase architecture. RNA binding is the existing, more specific assignment. Target sequence analysis retains the ATP-binding and DEAH motifs, although the broad RNA-binding conclusion does not depend on resolving every aspect of catalysis or YTHDC2-family phylogeny.

The target is a near-identical CSN5 splice product retaining the complete JAMM catalytic domain, and primary evidence establishes its COP9 deneddylation mechanism. The only experimental support for a synaptic-vesicle location in this ortholog group is a human COPS5 annotation from a mammalian study of the CSN complex in neuroblastoma cells and brain synaptosomes, whose abstract describes CSN4 binding and a role in synaptic protein stability rather than CSN5 residence in vesicles, and whose full text is not cached. Insects have no reported CSN5 synaptic localization, and no PAINT or curated phylogenetic transfer carries this term to the fly protein. The fly gene review records only nucleus and cytoplasm locations. The donor label from Xenopus Q6GLM9 is therefore not validated for the target, and no fly evidence refutes it. The term is absent from the GOA of both fly CSN5 products.

The PC isoform preserves the complete cyclin domain of characterized Drosophila CycA. Primary work on the CycA–Cdk1 system demonstrates control of CDK substrate recognition through the cyclin hydrophobic patch. The one-residue N-terminal isoform difference lies outside that domain, supporting the regulator activity. The canonical product of the same gene, P14785, already carries this exact term in GOA by experimental evidence (accepted as its core function in the gene review), so the prediction is not novel and CNN applies; actual inclusion in the model training set is not established.

The term is correct for fly cyclin A but less specific than annotations the gene already carries. Mitotic cell cycle phase transition is the parent of both G2/M transition of mitotic cell cycle (GO:0000086) and G1/S transition of mitotic cell cycle (GO:0000082) (is_a, verified in OLS4). The canonical product P14785 already carries GO:0000082 (IBA and IMP, kept as non-core), and the re-reviewed gene identifies G2/M transition as the core process, supported by cycA mutant and cycA;cycB double-mutant phenotypes. Isoform C retains the complete cyclin domain, so the gene-level evidence applies to this product.

ABC-type transporter activity is supported by the complete ABCB6-type architecture of Hmt-1 (TMD0, ABC transmembrane domain and ABC nucleotide-binding cassette) and by functional rescue of the fission-yeast hmt-1 cadmium-hypersensitivity phenotype by DmHMT-1, which acts without transporting Cd-phytochelatin complexes; the human ortholog ABCB6 is an ATP-driven porphyrin/heme transporter and heavy-metal efflux pump. The identical term GO:0140359 is already annotated to Hmt-1 (InterPro2GO IEA) and the gene review accepts it as the core molecular function, the precise level the cadmium evidence allows because the transported cadmium species is unresolved. The more specific ABC-type heme transporter activity is also carried (IBA, eumetazoan ABCB6 PAINT node), but it does not describe the cadmium role, so the prediction is not materially less precise than the curated core function. Not novel.

Full-length Henna (P17276) participates in aromatic amino acid metabolism by catalysing phenylalanine 4-hydroxylation. Hn-PD (178 residues, identical to P17276 275-452, initiating at internal Met275) lacks the ACT domain and the regions corresponding to the human PAH substrate-responsive loop 131-155, pterin-binding loop 245-250 and cofactor-stacking Phe254; only 149 of 307 catalytic-domain positions remain, although the iron ligands are retained. PD is therefore unlikely to perform the hydroxylation step itself. That does not refute participation in the process. PD retains the C-terminal part of the Henna sequence, the native product has not been assayed, and a non-catalytic contribution, such as association with or modulation of full-length Henna, is neither shown nor excluded. The missing catalytic architecture accounts for the catalytic route only, and there is no positive evidence against involvement, so the call is uncertain. Novelty is not at issue, as the term is an existing electronic annotation on E8NH57.

The exact native PD isoform lacks the PAH substrate-responsive loop, pterin-binding loop and cofactor-stacking Phe site, established by human-to-full-length-Hn mapping and their absence from PD. Primary structural and mutagenesis studies identify these as reaction-center features; only 149 of 307 aligned catalytic-domain residues remain. The three iron ligands and Glu cofactor contact are retained, but cannot supply the missing substrate/cofactor architecture. This refutes transfer of conventional pteridine-dependent hydroxylase chemistry, despite overlap with an existing electronic annotation. The conclusion concerns intrinsic catalysis, not all possible process participation or a supposed error in the native transcript sequence.

Protein phosphorylation is directly established by experiments showing that fly Ik2/IKKepsilon phosphorylates its substrates Spn-F, Nuclear fallout and DIAP1, and the Q9V3Y8 protein retains the complete kinase domain. GOA carries experimentally supported protein serine/threonine kinase activity (GO:0004674) for this gene, but no protein phosphorylation process term or descendant of it, and the molecular-function term does not entail the process term in the ontology, so the process annotation is novel. It does not entail every mammalian TBK1 downstream pathway, particularly interferon production.

Microtubule binding is directly supported by CH-domain experiments, full-length localization and the bundling phenotype. The same broad term already appears in GOA. Although GOA also contains plus-end binding, primary evidence shows that Mst27D binds along the lattice rather than sharing EB1’s preference, so the unsupported narrower annotation is not grounds for LSP. ProtNLM correctly captures the established broad activity; the gene review refines the IEA row to the more specific microtubule lateral binding (GO:0099609), which is not yet in GOA. The OpenScientist audit also retained GO:0008017 while identifying plus-end, spindle and dynamics terms as EB1/MAPRE-family IBA carry-over.

Conventional metalloendopeptidase activity is refuted by specific disruption of the M13 catalytic apparatus. Exact-sequence alignments map human H584/E585 to Nepl19 Q511/Q512, replacing a zinc ligand and catalytic glutamate; the downstream zinc-binding/catalytic region remains aligned. Two MAFFT strategies agree, and the active mouse control preserves all five sites. The published Nepl classification and existing NOT peptidase annotations agree with this independently reproduced sequence finding. The closest tandem paralogs in the 98F1 cluster keep HExxH, so the loss is specific to Nepl19 and family-level transfer of the catalytic term does not apply. The gene already carries accepted NOT peptidase activity annotations, which the predicted child term contradicts. This is pseudoenzyme overannotation, without a claim that the protein lacks a noncatalytic role.

The generic ligand-gated channel claim is supported by the target's Cys-loop architecture (signal peptide, Neur_chan_LBD ligand-binding domain with an intact Cys-loop disulfide, four C-terminal hydrophobic segments) and by the published Cys-loop channel-family tree, which includes Dmel NTR. The exact term is already in the exact-accession GOA record (IEA, InterPro2GO), which the gene review accepts as the best-supported molecular function, so the prediction is correct but not novel. No PubMed literature on NtR (CG6698) function was found. NtR lies outside the identified nicotinic receptor groups, so this broad result does not validate the narrower acetylcholine-gating or calcium-permeability annotations. The OpenScientist audit reached the same split, retaining broad channel terms and downscoping the acetylcholine, calcium and cholinergic assignments as untested computational propagation.

Pld hydrolyzes phosphatidylcholine to phosphatidic acid and choline, a lipid-catabolic reaction demonstrated for the Drosophila enzyme, which loses activity when a conserved catalytic residue is mutated. The prediction is therefore correct, but the same gene already carries the more specific is_a descendant GO:0009395 phospholipid catabolic process, by PAINT IBA on the canonical product B7YZT5 (node PTN004499769), and GO:0006654 phosphatidic acid biosynthetic process on this product. Lipid catabolic process adds no information beyond those annotations. Both annotated HKD catalytic regions are present in A4UZ54.

Immunolocalization in cellularizing embryos places Pld in the cytoplasm and on small cytoplasmic vesicles. The gene already carries GO:0031410 cytoplasmic vesicle by IDA, which is part_of cytoplasm, so the prediction is correct but less precise than the existing supported annotation rather than an erroneous cytosolic default.

The pseudouridine-synthesis prediction is correct. The published alignment of the exact Q9VKV0 protein shows that the conserved motif I residues and the residues needed for catalysis are retained, so the protein can be expected to carry out the RluA-family isomerization. The gene already carries this exact term in GOA, so the prediction is not novel. It is broader than the tRNA pseudouridine synthesis process supported by the gene's RPUSD2 subfamily placement, but that process is not yet an existing annotation on the gene, and the rRNA process term it does carry is an over-propagation; LSP therefore does not apply. The fly protein's endogenous substrates remain untested.

The fly study detects TyrRS-HA in S2-cell supernatant after serum deprivation, with membrane-integrity and tubulin controls excluding release by cell lysis. That shows TyrRS is released by a starvation-exposed cell, but TyrRS is the secreted cargo there; release does not show that the protein performs a step of a starvation response, and the gene review marks the broader ARBA-derived response-to-stress annotation as over-annotated on the same grounds. The participatory mechanism that would justify the term, starvation-induced nuclear translocation of YARS1 and PARP1 activation, is documented for the vertebrate SF43 ortholog but was not propagated by PAINT and has not been tested for fly TyrRS. No fly study links TyrRS to a starvation response (PubMed search of Drosophila TyrRS with starvation or nutrient terms returned no records). Available evidence neither validates nor refutes the prediction.

The exact predicted term, GO:1905594 resveratrol binding, is obsolete, replacement-less and explicitly out of scope for evolved molecular functions. OpenScientist found the fly tyrosine active-site pocket conserved with the human YARS1 pocket that binds resveratrol, so in-vitro binding by fly Q9VV60 remains plausible by homology but unmeasured. The biological prediction therefore cannot be validated or refuted with available evidence, and the obsolete xenobiotic-specific GO term should not be curated even if an in-vitro fly-binding assay is eventually positive.

A microtubule-based process is supported by the intact beta-tubulin architecture and primary experiments on both internal splice variants. Both forms partially rescue loss of the major betaTub56D paralog; the detailed stabilization and cold-induction results specifically concern the major exon-4B form. Exact correspondence of Q8MST5 to the paper’s exon names remains unresolved, but this does not undermine the broad structural microtubule role. Existing microtubule-cytoskeleton organization is a more precise supported process, so the prediction is LSP.

Nucleocytoplasmic transport is supported by the direct fly Imp13-Mago-Y14 import complex and the conserved bidirectional importin-13 cycle. It is broader than the existing protein-import and export annotations.

Cadmus directly carries Mago-Y14 in nuclear import. Intracellular protein transport is correct but less specific than the supported nuclear-import annotation already in GOA.

The prediction is not in GOA, which carries no molecular-function annotation for dati. A general DNA-binding transcription factor role is supported by the nuclear six-finger C2H2 architecture and the primary study's classification of DATI as a Krueppel-like zinc-finger transcription factor. That study establishes requirements in neuronal precursors and postmitotic neurons but does not measure activation of any target promoter, and no reporter or transcriptome assay for DATI exists. The related Rotund, Squeeze and ZNF384 proteins are in different PANTHER subfamilies or families, and PTHR16515 is heterogeneous with no PAINT node placed over dati, so neither the activating direction nor RNA polymerase II specificity can be transferred. The claim cannot be validated or refuted.

The prediction is not in GOA, which carries no DNA-binding or other molecular-function annotation for dati, so it would be novel. DATI has six C2H2 zinc fingers detected by PROSITE and is nuclear in neurons, which makes sequence-specific DNA binding plausible and supports a general DNA-binding transcription factor activity inferred from domain architecture. However, no DNA-binding assay, binding motif or target cis-regulatory region has been reported for DATI. The relatives with DNA-binding data (ZNF384/Nmp4) belong to a different PANTHER family, and PTHR16515 combines heterogeneous zinc-finger regulators without a PAINT node covering dati, so the RNA polymerase II cis-regulatory specificity cannot be justified by phylogenetic transfer. The specific claim remains unverified.

Correct but not novel. EMSAs show that Jumu binds a Forkhead (FkhS-type) site in the Nidogen mesodermal enhancer in a sequence-specific manner, and Jumu cooperates with CHES-1-like to repress that enhancer in the heart. GOA already carries the predicted term itself (IEA, InterPro) and the more specific descendant GO:0000987 cis-regulatory region sequence-specific DNA binding (IBA, accepted in the gene review and also proposed there as the refinement of the IDA cis-regulatory region binding row); GO:0000987 is_a GO:0043565 (OLS4 ancestry), so the prediction is less precise than an existing supported annotation.

Loqs-PF retains all three domains of the experimentally characterized PB miRNA-processing cofactor, differing by one linker-region lysine. Primary experiments demonstrate Loqs-dependent regulation of miRNA-target expression. Existing annotations include the more informative pre-miRNA-processing function, as well as the broad predicted term itself. LSP therefore reflects a correct but less precise description; ARBA agreement is not treated as validation. The mammalian pathway wording in the separate narrative prediction does not invalidate this broad GO claim.

Cell-junction localization is biologically supported by direct immunolocalization of Metro to the subsynaptic reticulum (SSR) of type I boutons at larval neuromuscular junctions, where it stabilizes junctional DlgS97 with DLin-7. The gene already carries GO:0071212 subsynaptic reticulum by IDA (an is_a/part_of descendant of cell junction via synapse, checked in OLS4) and GO:0005911 cell-cell junction by IBA, so the prediction is correct but less precise. The identical cell-junction term in GOA rests only on an ARBA rule, which the gene review refines to SSR.

Intracellular protein transport is directly supported by the loss of activated ERK nuclear import in moleskin mutants and rescue by wild-type DIM-7. A second primary study identifies Moleskin association with snurportin/snRNP cargo and an snRNP-import defect. The existing protein import into nucleus annotation is more specific than the predicted general intracellular transport.

Orb is an experimentally established CPEB-family RNA-binding translation regulator, and the same GO term already occurs in the selected record. Direct RNA-target profiling and germline translational studies support correctness independently of that electronic annotation. Q8IMZ2 is shorter at the N-terminus but retains the complete C-terminal RNA-binding machinery, making transfer of this broad function reasonable. The prediction adds no novel activity.

Spliceosomal processing and regulation of alternative mRNA splicing support a role in gene expression; existing GOA contains the more precise splicing assignments.

RNA binding is supported by the intact KH domain and spliceosome association. Existing mRNA-binding annotation is more informative.

Sns is required for myoblast fusion in body-wall muscle formation and for adhesion and fusion of garland nephrocytes into binucleate cells. Both myoblast fusion (GO:0007520) and garland nephrocyte differentiation (GO:0061321) are is_a descendants of cell differentiation (checked in OLS4) and are carried by the gene with IMP evidence; the exact term is also already in GOA for this accession. The prediction is correct but less precise than existing supported annotation.

Sns-mediated preferential adhesion maintains ommatidial separation in the developing eye, and Sns contributes to spacing of sensory bristles. The gene carries compound eye morphogenesis (GO:0001745, IMP) and chaeta morphogenesis (GO:0008407, IGI), both descendants of anatomical structure morphogenesis (checked in OLS4); the exact term is also already in GOA for this accession. The prediction is correct but less precise than existing supported annotation.

Sns is a transmembrane cell-surface adhesion receptor that is enriched at the membrane at myoblast fusion sites and localizes to the nephrocyte plasma membrane and nephrocyte diaphragm. The gene carries plasma membrane (GO:0005886, IDA and IBA), an is_a descendant of membrane (checked in OLS4), and nephrocyte diaphragm (IDA). The isoform-specific extra hydrophobic segment does not undermine membrane association. The prediction is correct but less precise than existing supported annotation.

D5SHU8 is the 299-residue ttv-PC product, identical to Ttv residues 462-760, so it contains only the C-terminal GT64 (GT-A) domain. It keeps the DxD motif and the active-site Asp equivalent. The same term is already carried by D5SHU8 as an InterPro2GO electronic annotation, so the prediction is not novel; the gene review marks that row over-annotated for this product. Family transfer is justified at the domain level, because Ttv and human EXT1 share a PANTHER subfamily. However, the human EXT1 GT-A domain is described as weakly active in the EXT1-EXT2 co-polymerase, and mutating it gave only minor activity changes in one study. A second study found reduced GlcNAc-T activity after EXT1 GT-A DxD mutation, so the domain may contribute activity inside the intact heterodimer. All of these measurements were made on full-length protein. ttv-PC lacks the signal-anchor helix and the GT47 domain and has no predicted signal or transmembrane segment, so it is not expected to reach the secretory-pathway acceptors or partner Sotv. Neither activity of the isolated domain nor its absence has been tested, and a PubMed search found no study of this isoform; UNC is retained.

The deposited sequence is a short RCC1-repeat protein with RCC1/BLIP-II fold assignments and no HECT or other ligase catalytic module. An RCC1 region from a larger HERC-like protein does not carry the catalytic activity of that protein's separate HECT domain. The prediction of intrinsic ligase activity is therefore incompatible with the available domain architecture. Sequence comparison of all products of LOC117183218 (the longest is 282 aa) finds no HECT domain in any of them, so the refutation holds at the gene level; the gene has no GO annotations, and the gene review records no established molecular function.

The target contains Exportin-5-specific domain and subfamily assignments across a full-length transport-receptor architecture. Experiments in Drosophila establish Exportin-5-dependent export of nuclear pre-tRNAs to the cytoplasm (PMID:39098529), and heterokaryon assays show that human Exportin-5 shuttles between nucleus and cytoplasm (PMID:11777942), supporting cytoplasmic residence during the conserved transport cycle. This is a family-based localization inference and does not imply exclusive cytoplasmic localization. Cytoplasm is already present in cached GOA and UniProt.

ALKAL2 is a secreted cytokine; the human ortholog is experimentally secreted and active when supplied as conditioned medium, and the donkey protein has a predicted N-terminal signal peptide and is 86% identical to human across its full length. The exact term is already in donkey GOA (UniProt subcellular-location mapping, accepted as a core location in the gene review), so the prediction is correct but not novel. Human ALKAL2 also associates with the cell membrane once bound to ALK, but extracellular region is the appropriate primary location for the ligand.

Caspase cleavage of human APP releases the C31 peptide, a potent inducer of apoptosis (PMID:10742146), and the gamma-secretase-released intracellular domain acts as a positive regulator of apoptosis (PMID:12214090); Swiss-Prot summarizes both as enhancers of neuronal apoptosis. The donkey intracellular domain, including the caspase site region and YENPTY motif, is identical to human, so the transfer is justified within the APP orthology group. No apoptosis term of any kind is in the donkey GOA record or the donkey gene review (human APP carries apoptotic process as an accepted IEA), so the prediction is novel for this gene. The role is fragment-mediated and non-core, and the evidence is for induction rather than execution, so positive regulation of neuron apoptotic process would be the more precise term.

After alpha-secretase cleavage the soluble APP ectodomain is released into the extracellular space and amyloid-beta is secreted; human APP has IDA support for the extracellular region. The exact term is already in donkey GOA (combined IEA). Correct but not novel.

The gamma-secretase-released C-terminal fragment (AICD) is located in both cytoplasm and nucleus of neurons (human EXP, PMID:11544248). The exact term is already in donkey GOA. Correct for a fragment of the gene product, though not for full-length APP; not novel.

Human ATG3 is experimentally cytoplasmic and acts as a soluble E2 that is recruited to phagophore membranes; the donkey protein is 98.4% identical with the catalytic cysteine and membrane-sensing N-terminus conserved, so cytoplasm is correct. Donkey GOA already contains cytosol and phagophore assembly site (TreeGrafter), both part of cytoplasm in GO, so the prediction is less precise than existing annotation. Training-set membership is unknown.

ATG4C is a C54-family cysteine protease that cleaves ATG8 proteins, and the donkey protein conserves the catalytic nucleophile. Proteolysis is therefore correct, but the donkey GOA record already carries protein processing (a child of proteolysis) and cysteine-type endopeptidase activity, so the prediction is less precise than existing annotations.

All four mammalian ATG4 paralogs delipidate membrane-anchored ATG8 proteins in reconstituted assays, and human ATG4C displays stronger delipidation than priming activity in the curated human UniProt function statement. The donkey protein is a full-length ortholog with the catalytic triad conserved. The donkey GOA record has the corresponding molecular function (protein-phosphatidylethanolamine deconjugating activity) but not this process term, so the prediction adds a correct, target-novel process annotation.

ATG4C acts in autophagy by priming and delipidating ATG8 proteins, so the term is correct. The donkey GOA record already has more specific autophagy annotations (autophagosome assembly, aggrephagy), so the prediction is less precise than existing annotations.

Dynactin binding is a true activity of the BICDL1 gene. Mouse BICDR-1 co-immunoprecipitates the major dynein and dynactin subunits, and in the cryo-EM structure of the dynein-dynactin-BICDR1 complex the BICDR1 coiled coil, modelled over mouse residues 105-392, lies along the dynactin filament. The term is already carried in GOA by the canonical product of the same donkey gene (A0A9L0K8K5), so it would not be novel. This accession, however, is a non-canonical splice form that lacks a 49-residue central coiled-coil segment (human Q6ZP65 residues 255-303, SEG_CORE) present in human, mouse and every other donkey product. That segment falls inside the coiled-coil stretch that contacts dynactin in the structure. The accession keeps the N-terminal CC1 box, but no experiment shows whether an adaptor with this shortened coiled coil still spans and binds dynactin. Whether this product retains the dynactin-binding interface is therefore unresolved.

Endogenous mouse BICDR-1 localizes around the gamma-tubulin-labelled centrosome, where it holds Rab6 secretory vesicles, and this localization depends on dynein and dynactin, since knockdown of either largely abolishes it. The term is already carried in GOA by the canonical donkey product A0A9L0K8K5, so it would not be novel. This accession is a non-canonical splice form that lacks the central coiled-coil segment (human Q6ZP65 residues 255-303, SEG_CORE) lying within the coiled coil that contacts dynactin. Because pericentrosomal targeting requires dynein-dynactin recruitment, and it is unresolved whether the shortened coiled coil of this product still engages dynactin, it cannot be determined whether this product localizes to the centrosome.

Mouse BICDR-1 is a Rab6 effector; full-length protein and its C-terminal region bind GTP-locked Rab6A but not GDP-locked Rab6A, and Lys512 is required for this binding. The donkey product is in the same PANTHER subfamily and keeps the C-terminal BICDL1/2-like region, so the transfer is a justified orthology inference. Small GTPase binding is the most specific applicable GO term used for this activity in the reviews. The term is already carried in GOA by the canonical donkey product A0A9L0K8K5 and is a core function of the target's gene review, so the prediction is correct but not novel.

The only experimental evidence linking BICDR-1 to neurites is an overexpression phenotype in cultured hippocampal neurons, where sustained BICDR-1 expression reduces neurite outgrowth by retaining Rab6 secretory vesicles in the pericentrosomal region; the authors conclude that BICDR-1 inhibits neuritogenesis early in differentiation. BICDL1 therefore does not perform neurite development; at most it negatively and indirectly regulates it through its vesicle-transport function. The positive process term matches the mouse IMP annotation that the mouse gene review modifies to negative regulation of neuron projection development, and the canonical donkey product's review marks the same term over-annotated on these grounds. The prediction reproduces that mis-signed source annotation rather than the supported role.

Human CADM1 is experimentally localized to the cell membrane as a single-pass type I protein, and the donkey sequence conserves the signal peptide, Ig-like ectodomain and transmembrane helix, so membrane localization is well supported. Donkey GOA already holds the more specific plasma membrane (GO:0005886) annotation, which is also the GO term named in the ProtNLM2 evidence record. The prediction is correct but less precise than an existing annotation.

GO:0017018 is defined as catalysis of phosphomyosin + H2O = myosin + phosphate, an activity carried out by the PP1c-MYPT1 myosin phosphatase holoenzyme of the PPP family. DUSP5 is a cysteine-based dual-specificity phosphatase of the PTP superfamily whose characterized substrate is ERK; human DUSP5 inactivates ERK2 but not JNK or p38, and its activity is highest toward ERK1. The donkey protein conserves the catalytic Cys263 and is 94.8% identical to human over its full length, so it is expected to share this ERK-directed specificity. No evidence links DUSP5 to myosin dephosphorylation, and the donkey GOA already records the correct MAP kinase tyrosine/serine/threonine phosphatase activity. The one study a PubMed search finds linking DUSP5 to myosin places it upstream, reporting that DUSP5 lowers myosin IIB activation indirectly by inhibiting ERK, which acts through MLCK, rather than by dephosphorylating myosin itself. The prediction assigns a substrate-specific phosphatase activity that belongs to an unrelated enzyme family (PPP-type myosin phosphatase), not to a paralog. No error-type value fits a training-data origin that cannot be shown here, so none is set.

Human ENTR1 (SDCCAG3) was identified in the VPS35 interactome, and its siRNA depletion reduces surface and total GLUT1, an SNX27-retromer recycling cargo, with minor effects on other cargo such as CD97 (PMID:25278552); the human gene review accepts endocytic recycling as an experimentally supported core process, while noting that the effect is cargo-selective and that ENTR1 also promotes lysosomal sorting of Fas (PMID:31308371). Transfer to donkey is an orthology inference, as donkey A0A8C4MKL2 (PTHR31259:SF3) and human ENTR1 (hominid-only SF6) are the single ENTR1 ortholog in each species, and the conserved C-terminal coiled-coil region shares 81% identity. The donkey gene review describes the recycling role in its description and core-function text but does not carry it as a GO annotation, because PAINT did not propagate this human term, and the term is absent from donkey GOA. The prediction is therefore a correct, novel annotation for this accession, with the caveat that it rests on human knockdown data rather than on the donkey protein.

Human ENTR1 co-purifies with VPS35 and colocalizes with the retromer cargo-selective complex on endosomes (PMID:25278552), but its association with retromer is mediated by the WASH complex (PMID:24980502), and the human gene review accepts the retromer term only with the colocalizes_with qualifier, as an accessory association rather than subunit membership. An unqualified retromer complex term implies membership, which no evidence establishes; the indirect, WASH-dependent association argues against a stable subunit role but does not exclude a peripheral one. The prediction is thus plausible as an association but neither supported nor refuted as complex membership. The term is absent from donkey GOA.

GO:0016310 phosphorylation is the process of introducing a phosphate group, so it applies to gene products that catalyze or directly carry out phosphate transfer. EPS8L2 has no kinase domain; its domains are PTB, SH3 and the EPS8 actin-binding effector region, and its characterized activities are F-actin association and stimulation of SOS1 exchange activity toward Rac. The ProtNLM2 evidence for this prediction is the Phosphoprotein keyword (KW-0597), which describes EPS8L2 as a substrate of phosphorylation rather than as an agent of it. The term is absent from donkey GOA, which instead carries actin binding, Rho protein signal transduction and ruffle-assembly terms, and the merged gene review describes EPS8L2 as an actin filament-binding adaptor that activates SOS1 Rac-GEF activity. The prediction is therefore refuted for EPS8L2.

Human FBXL20 binds PR55alpha (PPP2R2A), a regulatory subunit of the PP2A serine/threonine phosphatase, in vitro and in cells, and promotes its ubiquitination and degradation (PMID:34731788). The donkey product shares the canonical 422-residue F-box/LRR/CAAX body with human FBXL20 (same PANTHER subfamily PTHR13318:SF47), so the substrate-binding LRR domain is intact. The binding is a substrate-recognition interaction, so the term is correct but less informative than the adaptor activity the gene review records. Neither donkey GOA nor the human and mouse orthologs carry a phosphatase-binding term (human has only generic protein binding by IPI), so the prediction is novel.

Characterised SCF(FBXL20) substrates (RIM1 in mouse, VPS34 and PR55alpha in human) are ubiquitinated and degraded by the proteasome, which implies polyubiquitination. The only mention of monoubiquitination is the mouse SCRAPPER study stating that RIM1 monoubiquitination cannot be excluded, which was not tested. The term is neither validated nor refuted, and it is not in donkey GOA or the gene review.

In human cells FBXL20 knockdown induces autophagy flux and raises VPS34 and PtdIns3P, and DNA damage induces p53-dependent FBXL20 expression that drives SCF-mediated VPS34 degradation and inhibits autophagy (PMID:25593308). FBXL20 thus acts as a negative regulator of autophagy; negative regulation of autophagy would be the more precise term. Donkey GOA, the donkey gene review and the human and mouse orthologs carry no autophagy term, so the prediction is novel. Transfer rests on the shared FBXL20 subfamily and the intact F-box and LRR domains of the donkey product.

The experimental evidence (Bhd-null mouse cells and a FLCN-null human renal cancer line) shows that FLCN loss reduces TGF-beta-dependent transcription and alters expression of TGF-beta pathway genes, which supports at most regulation of the pathway, not participation in it. FLCN is a cytoplasmic and lysosomal Rag GAP with no reported role as a component of TGF-beta receptor signalling, and the effect is thought to be indirect, via TFE3/TFEB deregulation. The donkey GOA record carries positive regulation of TGF-beta receptor signalling, which the donkey review leaves UNDECIDED and the human review keeps as non-core; regulation of a pathway is not an is_a child of the pathway, so the prediction is not a less precise form of that term. The evidence neither establishes nor refutes direct involvement.

Human FLCN binds lactate dehydrogenase A directly and acts as an uncompetitive inhibitor (PMID:34381247, human IDA to this term; the human review prefers the more specific child L-lactate dehydrogenase inhibitor activity, GO:0160193, and treats it as non-core). The donkey protein is a full-length ortholog in the same PANTHER subfamily at 94.64% identity, and the term is absent from the donkey GOA record, which carries only GTPase activator activity as MF. The finding comes from a single laboratory and the inhibitory region was not mapped residue by residue, so the transfer rests on orthology.

FLCN loss abolishes mTORC1 phosphorylation of TFE3 and TFEB while leaving S6K and 4E-BP1 phosphorylation intact, because FLCN GAP activity on RagC is needed to recruit these substrates to the lysosome. The effect is therefore real but substrate-specific and mediated through mTORC1, and it is already captured more informatively by positive regulation of TORC1 signalling (GO:1904263), which the donkey GOA record carries and the donkey review accepts. GO:1904263 is not a descendant of GO:0001934 (OLS4 ancestry), so the prediction is not simply less precise. Neither human FLCN, FNIP1, RagC nor RagD carries this term or its descendants in GOA, indicating that regulators of this pathway are conventionally annotated to TORC1 signalling rather than to generic protein phosphorylation. The prediction is defensible but uninformative, and cannot be scored as clearly correct or incorrect.

Homozygous Flcn knockout mice die at 5.5-6.5 dpc, which shows that FLCN is necessary for embryonic development, not that it performs a step of it. The human FLCN review marks this term over-annotated on exactly these grounds; the mouse source is a phenotype-based IMP and the lethality is most plausibly a downstream consequence of unrestrained TFE3 activity. FLCN's activity is Rag GAP-mediated control of mTORC1, and no developmental-process mechanism has been shown. The term is absent from the donkey GOA record. The prediction converts a knockout phenotype into process participation, so it is scored as incorrect.

The only experimental basis is a knockdown study in mouse renal collecting-duct cells, in which FLCN or p0071 deficiency caused tight junction defects and E-cadherin mislocalization, and FLCN only partially co-localized with p0071 at junctions. This shows that FLCN deficiency perturbs junctions, not that FLCN participates in junction assembly, and the human FLCN review marks the term over-annotated on these grounds. No mechanism linking FLCN Rag GAP activity directly to junction assembly is known. The term is absent from the donkey GOA record.

The donkey protein carries a complete fork-head winged-helix DNA-binding domain that aligns without gaps to human FOXJ3 (98.1% identity), and mouse Foxj3 binds a conserved forkhead binding site in the Mef2c skeletal muscle enhancer. The prediction is therefore correct. The exact term is already present in donkey GOA (InterPro2GO), and donkey GOA also has the more specific RNA polymerase II cis-regulatory region sequence-specific DNA binding (TreeGrafter), which the ProtNLM2 evidence record cites. The prediction is less precise than existing annotation. Training-set membership is unknown. This accession is an exon-skipping isoform lacking full-length (A0A8C4PFN5) residues 211-253, a segment outside the fork-head domain with no mapped functional feature; the fork-head domain (78-173) is identical to the full-length model, so the product retains what this activity needs.

FOXJ3 is a forkhead transcription factor; the mouse ortholog activates Mef2c transcription through a conserved forkhead binding site in its enhancer (PMID:19914232). The donkey protein is 98.1% identical to human FOXJ3 with an intact fork-head domain, so the prediction is correct. The exact term is in donkey GOA (InterPro2GO) and the more specific DNA-binding transcription factor activity, RNA polymerase II-specific is in donkey GOA from TreeGrafter; the prediction is less precise than that. Training-set membership is unknown. This accession is an exon-skipping isoform lacking full-length (A0A8C4PFN5) residues 211-253, a segment outside the fork-head domain with no mapped functional feature; the fork-head domain (78-173) is identical to the full-length model, so the product retains what this activity needs.

Human GEMIN7 is a core SMN-complex subunit that binds SMN, GEMIN6 and Sm proteins (PMID:12065586) and sits at the centre of the complex with SMN and GEMIN8 (PMID:17178713); the donkey protein covers the complete human sequence at 88% identity. Membership of an Sm-like protein family complex is therefore correct. The exact term is present in donkey GOA (TreeGrafter), but donkey GOA also carries SMN-Sm protein complex, a more specific descendant of this term, so the prediction is correct but less precise than an existing annotation. Training-set membership is unknown.

The GPAT2 subfamily esterifies acyl-CoA onto the sn-1 hydroxyl of glycerol 3-phosphate, an O-acyl transfer; overexpressed mouse Gpat2 raises GPAT activity with arachidonoyl-CoA, and the donkey model conserves the acyltransferase domain and HXXXXD motif, so the predicted activity is correct in substance. Scored for this product, A0A9L0ILJ0 (718 aa) keeps the acyltransferase domain (199-333), the HXXXXD motif and both transmembrane segments of the full-length 800-residue model A0A8C4KXU5, and lacks only the conserved GPAT2 C-terminal region (about 85 residues replaced by a different tail), which is not the catalytic region, so the activity does not depend on a feature this product lacks. The term is obsolete in the current GO release (an unnecessary grouping class, replaced by GO:0016747, which is also in donkey GOA from InterPro2GO). Donkey GOA already carries the specific glycerol-3-phosphate O-acyltransferase activity (TreeGrafter), accepted as core in the gene review, so the prediction is less precise than existing annotation. The obsolete identifier is a defect of the emitted term, not of the biology. Training-set membership is unknown.

GPC3 regulates Wnt and hedgehog signaling as a GPI-anchored heparan sulfate proteoglycan at the cell surface; human GPC3 stimulates canonical Wnt signaling by binding Frizzled through its heparan sulfate chains. That mechanism needs entry into the secretory pathway, where the GPI anchor and heparan sulfate chains are added. A0A9L0J2B1 is an alternative-first-exon gene model whose first 43 residues are not homologous to the human N-terminus and lack the human signal peptide (P51654 SIGNAL 1..24). A Kyte-Doolittle and von Heijne-style n/h/c-region scan (signal_peptide_hydropathy.py) recovers the signal peptide and its cleavage after residue 24 in both human P51654 and the canonical donkey product A0A9L0JGR2, but finds no candidate h-region in A0A9L0J2B1; its most hydrophobic 7-residue window has a mean of 1.54, against 3.36 in the controls, and its longest apolar run is 5 residues. This is evidence against a classical secretory signal, so the known cell-surface mechanism is probably unavailable to this product. It is not positive refutation of the process term, however. The scan is a rule-based check rather than a trained predictor, the transcript itself is unconfirmed, and a contribution to signal regulation by the retained core protein through some other route is neither shown nor excluded. The call is therefore uncertain. The term is true of the GPC3 gene through its canonical product A0A9L0JGR2; donkey GOA already carries it and the more specific positive regulation of canonical Wnt signaling pathway (GO:0090263), so the prediction is not novel at the gene level.

HOXA2 binds DNA through its homeodomain, which is 100% identical between donkey and human, so the prediction is correct. Donkey GOA already contains DNA binding and, more specifically, RNA polymerase II cis-regulatory region sequence-specific DNA binding (GO:0000978) and sequence-specific double-stranded DNA binding (GO:1990837), both grounded in ortholog experiments. Because a supported, more specific annotation exists (GO:0000978 is a descendant of GO:0003677 and is accepted in the gene review, which marks DNA binding as too general), the prediction is less precise.

HOXA2 is a sequence-specific homeodomain transcription factor; mouse Hoxa2 has IDA support for repressing and activating RNA polymerase II targets. The exact term is already in donkey GOA (combined IEA) and is the core function accepted by the gene review; the donkey protein shares PANTHER subfamily PTHR45664:SF3 and an identical homeodomain with the characterised orthologs. Correct but not novel; the repressor child term GO:0001227 is also present in GOA, but HOXA2 also activates targets, so the general TF term is appropriate rather than less precise.

Nuclear localization is experimentally shown for mouse Hoxa2 (immunodetection in embryonic neural tube nuclei), and its regulated exit from the nucleus via the KPC2 interactor implies a nuclear steady state; human HOXA2 is curated as nuclear. The donkey protein is in the same PANTHER subfamily (PTHR45664:SF3) with an identical homeodomain, so the transfer is justified. The exact term is already in donkey GOA and is accepted as a core location by the gene review. Correct but not novel.

ISGF3, of which IRF9 is the DNA-binding subunit, activates RNA polymerase II transcription of interferon-stimulated genes, and human IRF9 is required for ISRE-driven reporter transcription. The donkey protein is in the same PANTHER subfamily (PTHR11949:SF26) with a conserved DNA-binding domain, which justifies the transfer. The exact term is already in the donkey GOA record and is accepted as the core process in the gene review, so the prediction is correct but not novel.

IRF9 binds the interferon-stimulated response element through its tryptophan-pentad DNA-binding domain, which is conserved in the donkey protein; human ISGF3-gamma binds DNA sequence-specifically. The term is correct and already in the donkey GOA record, but the gene review accepts the more specific descendant GO:0000978 (RNA polymerase II cis-regulatory region sequence-specific DNA binding; is_a ancestry checked in OLS4) on the basis of the PTHR11949:SF26 subfamily transfer and human data, and marks this term TOO_GENERAL. The prediction is therefore less precise than an existing supported annotation.

IRF9 supplies the sequence-specific DNA recognition of ISGF3 and acts as an RNA polymerase II transcription factor. The term is correct and already in the donkey GOA record, but the gene review accepts the more specific descendant GO:0000981 (DNA-binding transcription factor activity, RNA polymerase II-specific; is_a ancestry checked in OLS4) as the core function and marks this term TOO_GENERAL. The prediction is therefore less precise than an existing supported annotation.

Human KDM4C (JMJD2C) is an experimentally characterized H3K9 tridemethylase, and JMJD2 subfamily members convert H3K9me3 to lower methylation states. The donkey protein is 91% identical to human KDM4C and conserves every 2-oxoglutarate-, Fe(II)- and Zn(2+)-binding residue of the JmjC domain, so the activity transfers with high confidence. The exact term is already in donkey GOA (EC 1.14.11.66 mapping), so the prediction is correct but not novel. Training-set membership is unknown.

The donkey protein has a single N-terminal transmembrane segment (110-131) and human LNPEP is a single-pass type II cell-membrane protein that also resides in GLUT4-containing intracellular vesicles. Membrane is therefore correct. Donkey GOA already contains membrane and the more specific plasma membrane, so the prediction is correct but less precise than an existing annotation.

Donkey LNPEP conserves the M1 zinc ligands (H464, H468, E487) and the proton acceptor E465, and the human ortholog is a characterized zinc aminopeptidase that hydrolyzes oxytocin, vasopressin and angiotensin III. Metallopeptidase activity is therefore correct. Donkey GOA already contains this exact term (InterPro2GO) and its more specific descendant metalloaminopeptidase activity (TreeGrafter), so the prediction is less precise than an existing annotation.

Human LOXL3 is a protein-lysine 6-oxidase that oxidizes peptidyl lysine residues to allysine in collagens and elastin, so it directly performs peptidyl-lysine oxidation; the human gene carries this process by IDA, and mouse Loxl3 (same PANTHER subfamily) directly oxidizes fibronectin. The donkey protein keeps the copper-binding histidines and the LTQ-forming lysine and tyrosine, with an invariant catalytic domain. Donkey GOA has the molecular function (protein-lysine 6-oxidase activity) but no peptidyl-lysine oxidation row, and the MF and BP are in different GO aspects, so this is a novel correct process annotation rather than LSP or CNN.

Human MITF is a sequence-specific transcription factor that binds E-box and M-box promoter elements and regulates RNA polymerase II transcription of TYR and TYRP1; the donkey bHLH domain is 54/54 identical and the protein is 98.4% identical over paired residues. The predicted term is correct and is already in donkey GOA (InterPro2GO), but donkey GOA also has the more specific regulation of transcription by RNA polymerase II (GO:0006357, a descendant), and the gene review refines both rows to positive regulation of transcription by RNA polymerase II (GO:0045944) because human MITF-A activates the TYR and TYRP1 promoters. The prediction is therefore correct, not novel, and less precise.

Not in donkey GOA, nor on the sibling donkey product A0A8C4M4D0. Human NADK2 (same PANTHER subfamily PTHR13158:SF5) is a homodimer by crystal structure, SAXS and SEC-MALS (PMID:35868311), and an independent structure shows the same symmetrical homodimer (PMID:35733246). The element that mediates dimerization is extension 3/EMKA (human 325-367); deleting it gives a monomer. This accession retains that element: donkey 384-506 aligns to human 320-442. The segment it lacks (human 288-319) contributes S306 and A288 to the composite active site but is not the identified dimerization interface, so the product keeps the feature the prediction needs and the transfer stands.

Human NADK2 uses ATP as the phosphoryl donor (PMID:23212377). In the human structure the ATP site is formed by K76 next to the ATP triphosphate, S188 stabilising its phosphates, and the GGDG catalytic motif (PMID:35868311). All of these lie in human 1-287, which aligns to donkey 97-383 of this accession, and the GGDG motif is conserved (donkey 255-258). The segment this product lacks (human 288-319) contacts NAD+, not ATP, so the product keeps the ATP-binding features. ATP binding is not in donkey GOA, and NAD+ kinase activity is not a GO descendant of ATP binding, so this is a novel correct term rather than LSP.

For NADK2 this process is the phosphorylation of NAD+ to NADP+ (PMID:23212377). This accession provably lacks the segment equivalent to human 288-319: the donkey-human alignment aligns 0 of its 32 residues. The segment carries the TGSxAW NADK motif (human 314-319), whose W319 stacks the nicotinamide ring, and S306, which the second monomer contributes to the nicotinamide contacts (PMID:35868311). The product cannot form a complete NAD+ site of its own. The requirement of this segment for catalysis is inferred from the crystal structures; no mutagenesis of W319, S306 or the 288-319 segment and no activity assay of the isoform-2-equivalent product exist. The product keeps the GGDG catalytic motif, the ATP-site residues K76 and S188, the EMKA dimerization element and E379, which also contacts the nicotinamide in trans as E379-prime. It could therefore heterodimerize with the full-length product A0A8C4M4D0 and contribute residues to that partner's composite active site. This contribution is neither shown nor excluded. There is no positive refutation, and the missing segment does not fully account for the mechanism, so the call is uncertain rather than incorrect.

NDEL1 binds the interphase centrosome, the principal microtubule organizing center of animal cells, where it anchors microtubules. The exact term is present in the donkey GOA record (Ensembl projection from mouse), alongside the more specific centrosome annotation. The donkey protein is 97.1% identical to human NDEL1 across the full human sequence. Training-set membership is unknown.

NDEL1 is a soluble cytoplasmic and cytoskeletal protein, so cytoplasm is correct. The donkey GOA record already has more specific cytoplasmic locations (centrosome, microtubule organizing center, spindle, axon), each part of the cytoplasm, so the prediction is less precise than existing annotations.

NDEL1 is required for radial migration of newborn cortical neurons and, with DISC1, for regulation of neurite outgrowth; neither is an act of cell differentiation as such, and neither the human NDEL1 review nor the PTHR10921 family review assigns NDEL1 a differentiation role. The term is absent from the donkey GOA record. Literature linking Ndel1 to neural stem cell differentiation is indirect (lentiviral overexpression in cultured neural stem cells, and a mitotic kinetochore role in radial glial cell-cycle progression that affects neurogenesis secondarily); it shows an effect on the outcome, not participation in differentiation. The generic prediction can be neither supported nor refuted.

Human NUMB localizes to the cytoplasmic face of the plasma membrane, where it acts at clathrin-coated pits and accumulates when AAK1 is depleted (PMID:18657069); the donkey protein is 97% identical to human NUMB with a conserved PTB domain. The exact term is already present in donkey GOA (UniProt subcellular-location mapping), and donkey GOA also carries the more specific basolateral plasma membrane from the PANTHER node. The prediction is therefore correct but not novel. Training-set membership is unknown.

Human ORC5 is a subunit of ORC, which binds origins and is required to assemble the pre-replication complex that initiates DNA replication. This product, A0A9L0JAR4, keeps the AAA+ domain with the Walker A P-loop (GHTASGKT), the lid and the winged-helix with the DNA-proximal H398, matching human ORC5 at 402 of residues 1-420; it lacks only the last 14 residues of the canonical C-terminus, which are replaced by an alternative segment and have no known role in ORC assembly or replication. DNA replication is therefore correct for this product. Donkey GOA contains this exact term (combined IEA) and the more specific descendant DNA replication initiation (TreeGrafter), so the prediction is less precise than an existing annotation.

Human ORC5 is experimentally localized to the nucleus and chromatin (PMID:9765232; PMID:31160578 via UniProt), and ORC acts on nuclear replication origins. Human ORC5 has no annotated nuclear localization signal in its C-terminal 14 residues, the only canonical region this product A0A9L0JAR4 lacks; it retains the AAA+, lid and winged-helix regions that build the ORC2-5 core, so nuclear localization is expected for this product. The exact term is already in donkey GOA (combined IEA), so the prediction is correct but not novel.

The broad term cell cycle is not in donkey GOA, and no experimental GO annotation for human or mouse PARD3B places it in a cell-cycle process. Human PARD3B is described as a putative adapter in asymmetric cell division through its interaction with Inscuteable. The only direct cell-division evidence is a mouse knockout phenotype, in which Par3L-null intestinal epithelium has more mitotic cells and a higher fraction of defective spindles, accompanied by tumours. This is a loss-of-function phenotype in a tumour-prone tissue; it does not show which step of the cell cycle PARD3B performs or whether the spindle defects are direct, and the PARD3B-specific mechanism defined so far (LKB1 binding and restraint of its kinase activity) is a signalling role, not a cell-cycle step. The gene review removes the related TreeGrafter centrosome-localization term because it is a PARD3-derived transfer. The prediction is plausible but neither established nor refuted by the available evidence.

Human PARD3B colocalizes with TJP1 at epithelial tight junctions, and donkey GOA already contains bicellular tight junction (GO:0005923), which is a GO descendant of anchoring junction. The prediction is therefore correct but less precise than an existing supported annotation.

The input sequence fuses a KRAB-C2H2 zinc-finger segment (residues 1 to about 580, 64% identical to human ZNF786) to PDIA4; the ProtNLM2 evidence for this GO term is a weak phmmer match to mouse zinc finger protein 148 (Q61624), i.e. the zinc-finger part. Human PDIA4 is a 645-aa ER luminal disulfide isomerase with no zinc-finger region, the PDIA4 part of the donkey entry aligns to human residues 17-645, and the signal peptide needed for ER entry is missing, which together indicate a gene-model fusion rather than a real PDIA4 protein. A second donkey entry from the same Ensembl gene shows the same KRAB-type N-terminus. DNA binding is not a PDIA4 function (ERp72 is an abundant ER luminal protein), so the prediction is incorrect for this gene; the cause is the chimeric input sequence, not paralog confusion.

Full-length PIKFYVE contributes to phosphatidylinositol phosphate biosynthesis by synthesizing PtdIns(3,5)P2 and PtdIns5P through its C-terminal PIPK domain. The selected donkey protein is the product of a non-canonical transcript corresponding to human isoform 3, which ends at residue 548 and contains no part of the PIPK domain (human 1758-2084), so it cannot perform the catalytic step. That excludes the gene's main mode of participation, but it does not exclude every other mode. The short product also lacks the Cpn60_TCP1 (chaperonin-like) domain, which the PAS-complex study identifies as a major, not the sole, determinant of ArPIKfyve(VAC14)-Sac3(FIG4) association, and it lacks the complete FYVE finger that targets the enzyme to endosomes. It retains the N-terminal region and the DEP domain (donkey 277-352), whose contribution to the PAS complex or to lipid synthesis has been neither shown nor excluded. No study characterizes the short isoform, and endogenous human PIKfyve was detected as a single molecular species, so a binding or regulatory contribution is unsupported but not refuted. The prediction most likely transfers the full-length enzyme's output through the shared N-terminal and DEP region, and the term cannot be validated or refuted for this product, so it is scored UNC. At the gene level the claim is not novel, because the canonical donkey product A0A8C4L106 already carries phosphatidylinositol metabolic process and PI(3,5)P2 metabolic process annotations.

POLD2 is the essential B subunit of the replicative polymerase delta, and the donkey protein is 95.5% identical to human POLD2 across the whole human sequence, so DNA replication is biologically correct. The exact term is already in donkey GOA (InterPro2GO), and GOA also carries the more specific DNA strand elongation involved in DNA replication (TreeGrafter) and DNA-templated DNA replication (Ensembl Compara), both supported. The prediction is therefore less precise than existing annotation. The ProtNLM2 evidence record links it to the strand-elongation term. Training-set membership is unknown.

Human Pol delta, with POLD2 as an obligate subunit, performs about half of nucleotide excision repair synthesis and break-induced replication at broken forks (experimental evidence in the reviewed human record), and the donkey protein is conserved over the full length. DNA repair is absent from donkey GOA; the error-prone translesion synthesis row is a DNA damage tolerance process in GO, not a DNA repair descendant (QuickGO ancestor check). The prediction is a justified ortholog-based transfer that adds coverage to the target record. Novelty refers to target annotation coverage only; training-set membership is unknown.

Purified human RasGEF1A has guanine nucleotide exchange activity on small GTPases, specifically on Rap2 in one study and on H-, K- and N-Ras in another, so it acts at the activating step of small GTPase-mediated signalling whichever substrate is physiological. The donkey protein is in the same PANTHER subfamily (PTHR23113:SF172) as human Q8N9B8 and keeps the complete REM and catalytic Ras-GEF domains at 96% identity. The donkey gene already carries this exact term (InterPro2GO), which the gene review accepts as the core process term, so the prediction is correct but not novel. It is not less specific than a supported annotation, because the more specific Ras protein signal transduction is disputed for this gene and Rap2 signalling falls under a sibling term.

RNA binding is directly supported for the human ortholog, which sits in the same PANTHER subfamily (PTHR16105:SF2): the isolated human RBM41 RRM binds the U12 and U6atac snRNA 3' hairpins with low-micromolar affinity, and RNP1 point mutations or RRM deletion abolish its snRNA associations in cells. The donkey RRM is 78 of 79 residues identical to the human RRM, so the transfer is well justified. The exact term is already in donkey GOA, and donkey GOA also carries the more specific U12 snRNA binding (TreeGrafter), which the gene review accepts as the core function while marking RNA binding as too general. Because a supported, more specific annotation exists, the prediction is correct but less precise (LSP).

Not in donkey GOA, which has the broader polyamine metabolic process; the donkey gene review refines that to polyamine, spermidine and spermine catabolic process and leaves putrescine acetylation by SAT1 as an open question. SAT1's established substrates are spermidine and spermine (also norspermidine and 1,3-diaminopropane); their N1-acetylation and oxidation generate putrescine rather than consume it. Purified rat liver SSAT, the inducible SAT1-type enzyme, was inactive with putrescine. Against this, a mouse study attributes N-acetylputrescine production in the gut to Sat1, though without direct enzymology on purified Sat1 with putrescine, and human UniProt maps SAT1 to the UniPathway step 'N-acetylputrescine from putrescine' (which the human gene review keeps as a non-core role). With direct substrate-specificity data negative and the positive claim indirect, the evidence neither validates nor refutes a role of SAT1 in putrescine catabolism.

COPII-coated vesicle budding is a core process of the SEC23A gene, whose canonical product is an inner-coat COPII subunit and the SAR1 GTPase-activating protein; Sec23p-Sar1p interaction is required for budding in yeast, and a disease variant of human SEC23A that poorly recruits Sec13-Sec31 inhibits vesicle formation in vitro. The term is already in donkey GOA for this gene (InterPro2GO mapping of the Sec23 C-terminal domain), so it is not novel. The scored sequence, A0A9L0IQB9, is an in-frame exon-skipping transcript product that lacks the 26 residues equivalent to human 554-579. It keeps the zinc finger and its four Cys ligands, the trunk domain with F382, the beta-sandwich domain and the gelsolin-like PPP-motif-binding residues, but the deletion falls inside the Sec23/Sec24 helical domain. That domain spans residues 518-616 in the canonical donkey product A0A9L0JMP6, while its Pfam match on A0A9L0IQB9 covers only 553-590. Whether a Sec23 with a truncated helical domain still folds, binds SEC24 and SEC31 and supports vesicle budding has not been tested, and there is no expression evidence for this transcript. Because the required feature, an intact Sec23 fold, cannot be confirmed for this product, the prediction is uncertain at product level.

Scored for the A0A9L0JV53 polypeptide, an alternative transcript model of donkey SLC16A11 (MCT11, PANTHER PTHR11360:SF80). Transmembrane transporter activity is true of the gene, since human MCT11 transports pyruvate with a proton and the full-length donkey model A0A8C4PS96 has the complete fold, but this product lacks the region of transmembrane helices 1-3 of the twelve-helix major facilitator fold. Its first 107 residues are an unrelated sequence with no counterpart of human TM1-TM3, and only residues 108-439 (human 116-447) are shared with the full-length model. A major facilitator transporter missing three helices of its N-terminal six-helix bundle lacks part of the translocation pathway, so this exact polypeptide is not expected to transport on its own. The prediction depends on a feature this accession lacks.

SLC27A5 activates bile acids and long- and very-long-chain fatty acids to their CoA thioesters (human data, PMID:10749848 and PMID:10479480), and the donkey protein retains the AMP-binding catalytic domain at 81% identity, so participation in lipid metabolism is correct. Donkey GOA already carries the more specific descendants long-chain fatty acid metabolic process, very long-chain fatty acid metabolic process and bile acid metabolic process, so the prediction is correct but less precise than existing annotations. Training-set membership is unknown.

Membrane is true of the SNAP23 gene, whose full-length product is a palmitoyl-anchored peripheral plasma-membrane protein, but this 97-residue product lacks a feature that membrane association requires. Donkey residues 1..89 match human SNAP23 1..89 and keep the Qb SNARE motif and all five linker cysteines; the product then ends in an 8-residue alternative-exon tail and lacks the region corresponding to human 90..211. That region contains proline 119, the DHHC17 palmitoyl-transferase recognition residue. In mast cells the P119A mutation alone shifts full-length SNAP23 completely to the cytosol, as does loss of the linker cysteines, so membrane anchoring needs both the cysteines and the downstream P119 motif. Without P119 this product is not expected to be palmitoylated by DHHC17 or to associate with membranes, so the prediction is scored as a product-level architecture mismatch. Membrane association of this product has not been tested directly.

The donkey protein has a Phobius-predicted C-terminal transmembrane anchor (264-288), the tail anchor expected for a syntaxin, so membrane is correct. Human STX3 isoform A, to which the donkey sequence corresponds over its full length, localizes to the apical plasma membrane. Donkey GOA already carries both membrane and the more specific plasma membrane, so the prediction is correct but less precise than an existing annotation.

The donkey protein conserves the t-SNARE (SNARE motif) domain, and human STX3 is the apical receptor that mediates fusion of apical vesicles, so vesicle-mediated transport is correct. Donkey GOA already contains this exact term (InterPro2GO) together with the more specific vesicle fusion and exocytosis (TreeGrafter), both descendants of the predicted term. The prediction is therefore correct but less precise than existing annotation.

Betaglycan exists both as a membrane-bound receptor and as a soluble form in serum and extracellular matrix; the soluble ectodomain is released by proteolytic shedding mediated by MT1-MMP (PMID:14672946), and the human gene review keeps extracellular region as a non-core location. The donkey protein conserves the signal peptide, ZP domain and transmembrane helix, so the same membrane-bound and shed forms are expected. The donkey GOA has plasma membrane and external side of plasma membrane but not extracellular region, so the prediction adds a correct location. It is a secondary location; the primary site of receptor function is the plasma membrane.

The donkey model keeps the signal peptide and a single transmembrane segment (172-194), and human TNFR1 is a single-pass type I cell-membrane and Golgi-membrane protein, so membrane is correct. GOA for this product carries membrane raft (kept as non-core in the gene review), and the canonical product A0A8C4MD74 of the same gene carries plasma membrane (accepted in its review) and Golgi membrane; all are GO descendants of membrane, so the prediction is correct but less precise than existing annotations.

The cytoplasmic death domain (311-396) is retained and aligns without gaps to the human TNFR1 death domain, which nucleates TRADD recruitment and downstream NF-kappaB and apoptotic signalling. Signal transduction is therefore correct for this protein even though its ligand-binding ectodomain is internally truncated (exon 3 skipped). The exact term is already in GOA for this product (InterPro2GO from the death domain) and is accepted in the gene review; the more specific TNF-mediated signaling pathway row is UNDECIDED there because ligand binding is uncertain, so CNN rather than LSP is used.

The term is already in donkey GOA, but only via a family-wide UniRule (GO_REF:0000104) applied to peptidase C19 members, so the overlap does not establish correctness and the prediction is not novel. Donkey USP36 sits in PANTHER subfamily PTHR24006:SF653 (ubiquitin carboxyl-terminal hydrolase 36) with its catalytic Cys/His conserved, so the human ortholog's characterized activity transfers. That activity runs against the predicted process. Human USP36 deubiquitinates nucleophosmin, fibrillarin and c-Myc and thereby stabilizes them against proteasomal degradation (PMID:19208757, PMID:25775507), and PubMed searches for USP36 with degradation or proteasome return substrate-stabilization reports only, with no report of USP36 acting in a step of ubiquitin-dependent catabolism such as proteasome-associated chain trimming. The re-reviewed gene review marks the GOA row for this term as over-annotated on these grounds, which refute the prediction. ProtNLM2 repeats the generic DUB-family process label rather than the protein's substrate-protective role (protein stabilization).

Autophagosome is not among the cod gene's GOA terms; its only related annotation is the broader ARBA vacuole term (autophagosome is_a vacuole per OLS4), so the prediction is novel and more specific. The re-review sequence tree places cod tbc1d14 inside the vertebrate TBC1D14 clade, separate from TBC1D12, which makes the PANTHER SF15 assignment a subfamily mis-routing; the cod protein retains the TBC1D14 architecture and has its strongest phmmer match to reviewed bovine TBC1D14. Human TBC1D14 has an IDA autophagosome annotation from PMID:22613832 (partial colocalization of endogenous protein with LC3), and PAINT placed the same term on the TBC1D14-clade node; the human gene review keeps both rows as correct but non-core. This is a justified orthology transfer of experimentally grounded evidence. The localization is peripheral and partial: TBC1D14 acts mainly from recycling endosomes and the Golgi to regulate autophagosome formation, and no fish experiment addresses it, so it should not be read as constitutive residence.

The gene has no GOA annotations, so the term would be novel, but nothing establishes it. The kinase-like region matches the eukaryotic protein kinase profile (PF00069) only below the gathering threshold, small-molecule kinase profiles (Kdo, APH) align to the same segment, and no characterized relative or phylogenetic node is available for transfer. The gene review asserts only generic kinase activity and explicitly does not assert protein phosphorylation. A PubMed search for FFUJ_06423 and S0EDH7 returned no publications. Whether the protein phosphorylates proteins, small molecules or nothing is unknown.

The gene has no GOA annotations, so the term would be novel, but ATP binding is not established. The conserved catalytic-loop Asp/Lys/Asn and DFG Asp are compatible with an active phosphotransferase, yet the N-lobe glycine-rich loop and beta3 lysine that bind ATP in canonical kinases are not detectable in S0EDH7 or its homologs. No ligand-binding data or characterized relative exist, and no publication on the gene was found in PubMed. ATP binding remains plausible but untested.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by Ensembl Compara transfer from human Q8N4X5, whose IDA rests on Xu et al. 2007. In human BEAS-2B airway cells XB130 siRNA reduced IL-6 protein, and in Xb130 knockout mice serum IL-6 stayed elevated after LPS, so XB130 signaling does modulate IL-6 output, in a context-dependent direction. The unsigned regulation term fits both observations. The horse model is in the same 1:1 ortholog subfamily (PTHR14338:SF4) and retains the N-terminal SH2/SH3-binding region through which XB130 signals. The human gene review keeps this as a supported, non-core downstream output.

The term is carried in GOA on the horse AFAP1L2 product A0A3Q2HBW2 (IBA from PTN002682769 and Compara IEA), both seeded by the human IDA from Xu et al. 2007, so it is not novel; the issue is correctness. The underlying experiment shows only that XB130 enhances Src-dependent SRE/AP-1 reporter activity. AFAP1L2 is a cytoplasmic adaptor with no DNA-binding domain or nuclear role, and the reporter gain is a distal readout of its Src-activating activity. The human gene review marks both the IDA and IBA rows over-annotated on these grounds, which equally refute the prediction for the horse ortholog. The adaptor's real contribution is captured by protein tyrosine kinase activator activity.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by IBA from PAINT node PTN002682769 and by Compara transfer from human Q8N4X5. Independent human studies show altered cell cycles after XB130 knockdown (Xu 2007) and G1 arrest when XB130 or its Tks5 interaction is lost (Moodley 2015), downstream of Src-PI3K-Akt signaling. The horse model sits inside the PAINT clade and the same 1:1 ortholog subfamily, so the inherited regulatory role is a justified transfer. The human review keeps the term as a supported but non-core upstream signaling role.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by IBA from PAINT node PTN002682769 and by Compara transfer, and the horse gene review records it as the first core function. Human XB130 co-expression activates c-Src, and endogenous depletion lowers c-Src activity and blocks serum/EGF-induced Src activation. The node is confined to the AFAP1L2 subfamily (PTHR14338:SF4), which contains the horse model, so the transfer is phylogenetically justified. The term correctly describes a kinase activator, not a kinase.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by Ensembl Compara transfer from human Q8N4X5. In human airway epithelial cells XB130 knockdown reduced IL-8 production, which supports the positive sign. The horse model is in the same 1:1 ortholog subfamily and retains the N-terminal signaling region. The human review keeps this as a supported, context-specific and non-core downstream output of Src signaling.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by Compara transfer from the human IPI. Human XB130 carries N-terminal SH2-binding phosphotyrosine motifs needed for Src signaling, and Tyr54 docks the p85alpha SH2 domains; the horse model retains the whole N-terminal region and the equivalent tyrosine (Tyr65). AFAP1L2 binds SH2 domains rather than containing one. The more specific PI3K regulatory subunit binding activity is proposed in the horse review but not yet in GOA, so the prediction is not scored as less precise.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by Compara transfer from the human IPI. The polyproline-rich N-terminus of human XB130 is bound by the fifth SH3 domain of Tks5, and deleting the region carrying the SH3-binding motif impairs Src signaling. The horse model retains the full N-terminal region (167/167 paired residues) within the same ortholog subfamily, supporting transfer of this binding activity.

The term is carried in GOA on the horse AFAP1L2 product A0A3Q2HBW2 (IBA from PTN002682769 and Compara IEA), seeded by a human IDA from Xu et al. 2007, so it is not novel; the issue is correctness. That experiment measures reduced IL-8 and IL-6 release from cultured airway epithelial cells, which is captured at the right specificity by the cytokine-production terms and does not show that XB130 does any of the work of an inflammatory response. The human gene review marks both rows over-annotated on these grounds. In vivo, Xb130 knockout mice show enhanced, not reduced, LPS-induced cytokine production and lung injury, and more inflammatory skin injury after TPA, so the cell-line result does not generalize to a pro-inflammatory role. The prediction propagates an over-broad annotation.

Not novel: GOA carries this term on the horse AFAP1L2 product A0A3Q2HBW2 by IBA from PAINT node PTN002682769 and by Compara transfer. Human XB130 depletion reduces EGF-induced Akt and GSK3beta phosphorylation and growth factor-induced Src activation, supporting a positive contribution to signaling downstream of the EGF receptor through its Src and p85 docking roles. EGFR is one of several kinase inputs XB130 couples to PI3K, so the human review keeps the term as non-core. The horse model is inside the PAINT clade, justifying the transfer.

Human ALDH7A1 is an alpha-aminoadipic-semialdehyde dehydrogenase with experimentally demonstrated activity on several additional aldehydes. Those results establish ALDH chemistry but do not establish glyceraldehyde-3-phosphate oxidation. The selected horse sequence retains catalytic residues but has an internal 18-residue deletion, further limiting an exact substrate transfer. No target GOA overlap or direct glyceraldehyde-3-phosphate assay was found in the inspected evidence, and PubMed searches for ALDH7A1 or antiquitin with glyceraldehyde(-3-phosphate) returned no reports. The kinetic survey of purified human ALDH7A1 found only low affinity for the related short-chain aldose glyceraldehyde, and non-phosphorylating glyceraldehyde-3-phosphate dehydrogenase (GAPN) activity is characteristic of a different ALDH family. The different canonical substrate alone does not prove the predicted reaction is impossible for a broad-specificity ALDH, so the prediction remains uncertain rather than receiving a confident paralog-error label.

Beta subunits regulate high voltage-activated calcium channel currents without forming the pore. Coexpression of human beta3 modulates voltage-dependent calcium channel activity (PMID:8119293), and rat beta3 coexpressed with a cardiac alpha1 increases peak current, shifts the voltage dependence of activation and accelerates activation and inactivation (PMID:7679112). The horse protein is 99.78% identical to human CACNB3 over the shared core and groups with the beta3 lineage (PTHR11824:SF8) rather than the other beta paralogs, so this beta3-specific evidence transfers. The horse gene review records the same role as calcium channel regulator activity in its core function. Horse GOA carries calcium ion transport, calcium ion transmembrane transport and channel-complex terms but no regulation-of-transport process term, so the prediction is novel for the target. It is a correct but general parent of the positive regulation of voltage-gated calcium channel activity that the evidence supports. Training-set membership is unknown.

The prediction is not in horse GOA for either CC2D2A product. Mouse Cc2d2a mutants show dextrocardia with situs inversus and atrioventricular septal defects in which atrioventricular-cushion cilia are absent, so the heart phenotype is real. It is, however, a downstream consequence of loss of cilia at the node and in the cushions: CC2D2A is a transition-zone and subdistal-appendage scaffold that performs no step of heart development. The mouse gene review marks the mouse heart development IMP as an over-annotation for this reason and states that it should not be transferred to orthologs, and no horse or PAINT evidence supports an organ-development role. The prediction is therefore refuted as a functional claim for the horse protein.

The horse CC2D2A gene already carries smoothened signaling pathway in GOA (Ensembl Compara IEA from mouse on the canonical product A0A5F5PUW7), and the horse canonical-product review keeps it as a true but non-core process. In mouse, Cc2d2a-null fibroblasts show strongly reduced ciliary Smoothened after SAG stimulation and null embryos lose Shh-dependent ventral neural-tube fates. A0A5F5PJ44 is a splice model of the same gene that retains the coiled coils and C2 domain at high identity to human CC2D2A, so the orthologous transfer is justified. CC2D2A contributes by gating ciliary membrane composition, not by any receptor or transducer activity.

The prediction is not in horse GOA. Cc2d2a loss in mouse leaves perinatal kidney tubules with few cilia, and a forward-genetic Cc2d2a line develops prenatal cystic kidney disease, but these are ciliopathy phenotypes of a protein that acts at the transition zone and mother centriole of essentially all ciliated cells; CC2D2A performs no step of kidney development. The mouse gene review marks the mouse kidney development IMP as an over-annotation on these grounds and advises against transfer to orthologs, and PAINT propagates no organ-development term in this family. The prediction is therefore refuted as a functional claim for the horse protein.

Neural tube closure is not in horse GOA for either CC2D2A product. The targeted mouse Cc2d2a null has an open neural tube, and the authors trace it to loss of neural-tube cilia and cilia-based Shh signalling; the mouse gene review keeps this IMP as a valid non-core process. The horse protein is an ortholog in the same CC2D2A subfamily with high identity across its coiled-coil and C2-domain scaffold, so the transfer is an orthology inference rather than a horse observation. The term describes a developmental consequence of the core ciliogenesis role, not a CC2D2A activity.

The cyclin-dependent kinase activity is supported by human CDK7 structural and biochemical work (PMID:15530371), the conserved horse catalytic Asp137 and ATP-binding residues, and 95.38% identity over the complete human reference; the horse gene is the CDK7 ortholog in PANTHER subfamily PTHR24056:SF0. The horse protein has an additional segment, but the catalytic domain is retained. The exact term is already in horse GOA, where the gene review accepts it as a core activity, so the prediction is correct but not novel. Training-set membership is unknown.

Human CDK7, in the cyclin H-MAT1 complex, phosphorylates CDKs, the RNA polymerase II CTD and Spt5 (PMID:16327805, chemical-genetic substrate identification), and the horse ortholog (PTHR24056:SF0) retains the catalytic domain and ATP-binding features at 95.38% identity. Protein phosphorylation is therefore a justified orthology transfer of a process the kinase itself performs. Horse GOA carries kinase MF terms but no protein phosphorylation term or descendant, and the gene review proposes none, so the BP term adds coverage; it is novel for the target annotation set, not a new biochemical discovery. Training-set membership is unknown.

Cholesterol 25-hydroxylation is a specific oxidoreductase reaction established for cloned human and mouse enzymes in PMID:9852097. The horse protein retains the conserved enzyme sequence and histidine-rich architecture. The prediction is biologically correct but less specific than cholesterol 25-hydroxylase activity already present in horse GOA. Training-set membership is unknown.

CH25H produces the oxysterol 25-hydroxycholesterol from cholesterol. That product-forming reaction supports lipid biosynthesis even though the product also represses cholesterol biosynthesis; inhibition of the latter pathway does not make this broad BP false. The exact target GOA term is present. This does not validate the separate zymosterol or C-4 methylsterol assertions. Training-set membership is unknown.

The cloned human and mouse enzymes are polytopic membrane proteins (PMID:9852097), and UniProt places human CH25H in the endoplasmic reticulum membrane as a multi-pass protein by similarity to mouse Ch25h. The horse sequence aligns almost full length to human and keeps the membrane-enzyme architecture. Membrane is therefore correct, but it is less specific than the endoplasmic reticulum membrane annotation the horse gene already carries by IBA, which the gene review accepts as the core location (the review refines the IEA membrane row to it). Training-set membership is unknown.

Kinase activity is incorrect for F7A4N8 on two independent grounds. First, the exact 174-residue input shares only an 84-residue N-terminal prefix with human CTDSP2, lacks the FCP1-homology catalytic domain and its DxDx(T/V) motif, contains none of the eukaryotic protein-kinase catalytic motifs, and has no confidently folded catalytic domain in its AlphaFold model. Second, the CTDSP2 family itself is not a kinase family; the full-length horse products (A0A9L0T5X1, A0A9L0R207) retain the DLDET motif and group with CTDSP2 orthologs in PANTHER PTHR12210:SF187, whose characterised member, human SCP2, is a small CTD phosphatase that preferentially dephosphorylates Ser5 of the RNA polymerase II CTD. No literature reports kinase activity for any CTDSP2 ortholog (PubMed search of CTDSP2 with kinase, horse or equine). The prediction therefore has no support for this product or for the gene.

Gene-expression regulation is correct for DNMT3A, which silences genes by de novo CpG methylation, and the horse model retains the ADD and catalytic domains. The gene already carries the accepted, more specific descendant GO:0044027 negative regulation of gene expression via chromosomal CpG island methylation, which states the mechanism; the predicted term is its ancestor (via epigenetic regulation of gene expression). The PWWP-region exon skip changes heterochromatin targeting but not methylation of other genomic sequences.

DNMT3A is nuclear, and mouse deletion mapping places its two NLS basic stretches around the PWWP domain, outside the exon skipped in this horse model, which also retains the chromatin-associated ADD and catalytic domains. The gene already carries the accepted, more specific descendant nucleoplasm (GO:0005654). Horse testis immunofluorescence is relevant but does not identify this accession or resolve compartments in the cached abstract.

Mouse Dnmt3L co-localizes with Dnmt3a and Dnmt3b in nuclei, and DNMT3L acts on chromatin as the activating partner of these de novo methyltransferases. The horse sequence aligns across the full human protein with an N-terminal extension. Horse GOA already carries nucleus (IBA and IEA, both accepted in the gene review), so the prediction is correct but not novel.

Loss of mouse Dnmt3L abolishes maternal methylation imprints and causes biallelic expression of maternally imprinted genes, and human DNMT3L represses transcription in an HDAC1-dependent manner. The horse protein is a full-length ortholog. Horse GOA carries negative regulation of gene expression, epigenetic (GO:0045814), accepted in the gene review and an is_a descendant of GO:0010468, so this broad prediction is correct but less precise than the existing annotation.

DUOX1's only established hormone-related role is supplying the H2O2 that thyroid peroxidase uses to iodinate thyroglobulin, which GO captures as thyroid hormone generation (GO:0006590), a term the canonical horse DUOX1 product F6YEX2 already carries. GO deliberately places thyroid hormone generation outside the biosynthetic-process branch; OLS4 shows GO:0006590 is not a descendant of GO:0042446, and its definition comment states that the term does not fall under the general GO definition for biosynthetic processes because thyroid hormones are formed by proteolysis of a larger molecule. No evidence links DUOX1 to the biosynthesis of any other hormone. The prediction therefore places a real thyroid contribution under a GO branch that excludes it, and the term is discordant with GO's definition of the process DUOX1 supports.

The mammalian DUOX1 catalytic system generates H2O2; the isolated human peroxidase-like domain binds no heme and has no intrinsic peroxidase activity. Producing the substrate for a separate peroxidase is not evidence that DUOX1 carries out H2O2 catabolism. The selected horse protein retains the homologous peroxidase-like region within the same DUOX1 subfamily as human DUOX1, and no independent catabolic mechanism is supported. The gene review removes the electronic peroxidase activity and cellular oxidant detoxification rows on these grounds. This prediction confuses an oxidant-generating enzyme with peroxide-consuming chemistry.

Thyroid hormone generation is a correct term for the DUOX1 gene, and the canonical horse product F6YEX2 already carries it. DUOX1 takes part in the process by generating the H2O2 that thyroid peroxidase uses. A0A9L0SQG9 lacks the segment corresponding to human residues 775-873, which removes EF-hand 1 and most of EF-hand 2; of the 15 human calcium-ligand positions it keeps 6, and 5 of those form the extracellular peroxidase-homology site, not the EF-hands. The catalytic machinery is intact, with all heme, FAD and NADPH contact residues retained. The EF-hand module is a calcium-gated activator, not the catalytic step itself; point mutations of its calcium-binding residues abolish activity of mouse DUOX1-DUOXA1, but human DUOX1-DUOXA1 in low calcium retains low basal H2O2 production, and no study has tested a DUOX1 lacking this segment. The participation mechanism is therefore not fully accounted for by the missing feature, and there is no positive refutation; the product could retain basal or calcium-independent oxidase output, or could be inactive or fail to mature with DUOXA1. Whether this exact sequence contributes to thyroid hormone generation is unresolved.

Calcium binding is directly observed in human DUOX1, whose EF-hand calcium occupancy switches the oxidase on, and horse DUOX1 already carries calcium ion binding on its canonical product F6YEX2, so the term is not novel for the gene. This splice product lacks EF-hand 1 and most of EF-hand 2, retaining 6 of the 15 human calcium-ligand positions. The retained ligands include the complete putative calcium site in the extracellular peroxidase-homology domain (human positions 109, 170, 172, 174 and 176), which the mouse DUOX1 structure places at the position of the lactoperoxidase calcium site, and the peroxidase-like region is fully present. Calcium ion binding is therefore supported for this exact sequence as well, by orthology within PTHR11972:SF75; what the deletion probably removes is calcium-regulated activation, not all calcium binding.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Endogenous human TCTEX1D2 colocalizes with WDR60 at the ciliary base, the site where IFT dynein is loaded onto trains; the human and mouse reviews accept this location as part of the core retrograde IFT context. Transfer within the subfamily is justified because the location follows from dynein-2 membership. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which ciliary-base localization, which follows from incorporation into dynein-2 through WDR60 depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Human TCTEX1D2 forms a WDR60-TCTEX1D2-DYNLT1/DYNLT3 subcomplex mapped by immunoprecipitation, and mouse endogenous tagged TCTEX1D2 co-precipitates dynein-2 intermediate chains; PAINT places the term at the Tctex family node ancestral to SF7. This is the core molecular function in the horse gene review and in the review of the canonical horse product A0A9L0S9P8. It is an orthology inference, not a horse experiment. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which this binding activity depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Human TCTEX1D2 colocalizes with WDR60 at mitotic spindle poles by immunofluorescence; the human and mouse reviews keep this as a supported but non-core location. The term is peripheral to the ciliary dynein-2 role, and rests on a single human imaging study. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which spindle-pole localization, which in human cells accompanies WDR60 depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Human TCTEX1D2-knockout cells and patient fibroblasts show retrograde ciliary trafficking defects, and Chlamydomonas Tctex2b loss destabilizes IFT dynein; the protein contributes structure to the dynein-2 tail, which is participation in the process. This is the core process in the human and mouse reviews, and the horse review's core function places the gene in retrograde IFT by orthology. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which participation in retrograde IFT as a structural dynein-2 tail subunit depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The evidence behind this term is loss-of-function: removing TCTEX1D2 impairs retrograde IFT and destabilizes the IFT dynein motor. That shows the protein is part of the machinery that executes retrograde IFT, not a modulator of a process carried out by others. The human and mouse reviews both correct the corresponding annotations laterally to GO:0035721 intraciliary retrograde transport on exactly these grounds, which also refutes the regulation term predicted here; the prediction reproduces the upstream regulation framing. The participation term GO:0035721 is predicted separately; it is true of the gene but is scored UNC for this alternative model.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Human TCTEX1D2 depletion caused ciliogenesis defects in one study, and in mouse the knockout disrupts sperm flagellum assembly (a descendant of cilium assembly) while somatic motile cilia appear normal. Cilium assembly is broad but true for the gene; no more specific term is annotated to the horse gene, so LSP does not apply. The human review keeps it as non-core because the effect is downstream of retrograde IFT. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which the cilium-assembly role, a downstream consequence of dynein-2 function depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Human TCTEX1D2 colocalizes with WDR60 at interphase microtubule-organizing centers (centrosomes), dependent on WDR60 and microtubules; the human and mouse reviews keep centrosome as a supported non-core location consistent with dynein-2 recruitment near the basal body. The gene-level evidence is a single human imaging study. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which centrosomal localization, which in human cells depends on WDR60 depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Mouse sperm fractionation recovers endogenous tagged TCTEX1D2 in the axonemal fraction and the authors conclude it is an axoneme component; tagged human TCTEX1D2 also localizes to the ciliary axoneme when overexpressed. The mouse review accepts axoneme as a core location for the flagellar role. Transfer within SF7 is justified, though the strongest evidence is from sperm flagella. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which axonemal localization as a dynein-2 component depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. Human TCTEX1D2 colocalizes with WDR60 at microtubule-organizing centers during interphase; the human and mouse reviews keep this as a supported non-core location. The gene-level evidence is a single human imaging study. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which MTOC localization, which in human cells depends on WDR60 depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

The horse protein is a one-to-one orthologue of human DYNLT2B in the DYNLT2B-specific PANTHER subfamily PTHR21255:SF7, with 91% identity over the aligned Tctex domain. TCTEX1D2 depletion in human cells increases cilia length, a typical retrograde IFT phenotype, which supports a role in controlling cilium assembly; the human and mouse reviews keep this term as non-core because it is a downstream consequence of the protein's dynein-2 subunit role rather than a dedicated regulatory activity. The term is of limited informativeness. The horse gene has no GOA annotation on any product, so the term is novel for it. Product-level assessment: ProtNLM2 was given A0A9L0SWY1, which keeps the Tctex domain signatures (Pfam PF03645, CDD cd21459) and the human Tctex strands at residues 4-11, 89-99, 104-113 and 119-126, but ends two residues after strand 119-126 and so lacks the C-terminal Tctex beta-strand (human residues 128-140), and it adds an approximately 117-residue disordered N-terminal extension. In the human DYNLT1-DYNLT2B heterodimer crystal structure (PDB 8RGI), most DYNLT2B residues within 4 angstroms of DYNLT1 lie in the retained strands 89-112, and only residues 137, 139 and 140 of the missing strand contact DYNLT1, but that strand belongs to the core beta-sheet of the fold. Whether this truncated product folds, heterodimerizes with DYNLT1 or DYNLT3 and binds WDR60, on which the cilium-length phenotype, a downstream consequence of dynein-2 function depends, is untested, so the prediction is UNC for this product although the term is correct and novel for the gene.

Recruitment of the PI4KA complex to the plasma membrane is the core process of EFR3A, and the horse EFR3A gene already carries GO:0072659 by IBA on its full-length product A0A3Q2GU27, so the term is not novel at the gene level. It is incorrect for this exact product, on the basis of positive experimental refutation rather than absence of evidence. EFR3A takes part in this process as the membrane anchor of the PI4KA complex: EFR3 proteins are the membrane localization signal of the complex, and in higher eukaryotes palmitoylation of the N-terminal Cys-rich motif is required for plasma membrane association (PMID:38746453; PMID:24360784). A0A9L0S4L8 is an alternative-first-exon product (N-terminus MIDKDGLVKADMEK..., no cysteine and no Gly2 myristoylation site) that aligns to human EFR3A at 98.24% identity downstream but lacks human residues 2-26, including the Cys6-Cys9 palmitoylation cluster. It retains the conserved basic patch, which is the only other targeting element identified, but in mammalian cells that patch is an additional, not a sufficient, signal: the C6S/C7S/C8S/C9S EFR3A mutant keeps the basic patch and the TTC7-binding scaffold yet is cytosolic, and the equivalent palmitoylation-deficient EFR3B mutant cannot recruit TTC7B and PI4KA to the plasma membrane (PMID:23229899). That mutant reproduces the defect of this accession, so the anchoring mechanism through which EFR3A contributes to the process is shown experimentally to fail without the missing motif. A contribution through incorporation into a complex anchored by another palmitoylated EFR3 copy has not been described. Training-set membership is unknown.

Human GEMIN5 binds the Sm site of snRNAs through its tandem WD40 domain and is required for SMN-complex Sm-core assembly. A0A9L0R5P7 is in the same PANTHER subfamily (PTHR46362:SF1), shares 89% identity with human, and keeps the complete folded WD40 module (human 3-722) with all 12 mutationally validated RNA-contact residues; its missing block (human 723-798) is a disordered linker outside the fold. The transfer is therefore justified, as an inference from the human ortholog. The gene already carries this term in GOA on its full-length product F6WDM0 (IBA from PAINT and Ensembl Compara IEA), and the gene review lists it as a core process of this product, so the prediction is correct but not novel.

The characterised activities of GEMIN5 are snRNA and cap recognition for snRNP assembly, SMN mRNA 3'-UTR binding and ribosome-associated translation control. Neither the human nor the horse gene review, the GOA files of either horse product, nor a PubMed search for GEMIN5 with ubiquitin, ubiquitination or E3 (three hits, none describing a GEMIN5 role in ubiquitination) supports participation in protein ubiquitination. GEMIN5 has no ubiquitin-ligase or ubiquitin-binding domain, but the absence of a catalytic domain would not by itself refute a noncatalytic role. No evidence supports or refutes the prediction.

Immunolocalization of human GEMIN5 shows a minor nuclear pool that colocalizes with SMN in gems, besides the predominant cytoplasmic pool, so nucleus is true of the gene (and less specific than the nucleoplasm and Gemini of Cajal bodies annotations the gene carries). For this product the call is unresolved. A0A9L0R5P7 lacks the block equivalent to human 723-798, which in the full-length horse product F6WDM0 includes the basic-residue cluster at 748-757 (PKPKKKKK), a lysine run of the kind that can act as a classical monopartite nuclear localization signal; the product keeps a shorter KKRK motif at its residue 765. No nuclear-import determinant of GEMIN5 has been mapped, and its nuclear pool could instead depend on co-import with the SMN complex, so it cannot be told whether this product keeps what nuclear localization requires.

Human GEMIN5 binds the m7G cap directly in pull-down assays, and a soaked m7GTP structure places the cap in the central pore of the second WD40 propeller, stacked by Tyr474 and contacted by Leu580 and Asn582. A0A9L0R5P7 keeps the complete folded WD40 module; the deletion lies C-terminal to it. A direct ungapped comparison of the downloaded sequences in this assessment shows the cap-site residues W286, F338, Y474, L580 and N582 identical to human. The transfer from the same PANTHER subfamily is therefore justified. The gene already carries this term in GOA on its full-length product F6WDM0 (Ensembl Compara IEA, accepted by its review), so the prediction is correct but not novel.

Human GEMIN5 binds the SMN mRNA 3'-UTR directly, and PAINT places this activity at node PTN005299808, so the term is true of the gene, which carries it in GOA on both horse products. For this product the call is unresolved. A0A9L0R5P7 keeps the WD40 module and the C-terminal TPR-like and RBS regions, but lacks the block equivalent to human 723-798, which includes the lysine-rich basic cluster (F6WDM0 748-757), a type of segment that commonly contributes to RNA contacts. The SMN 3'-UTR binding interface has not been mapped to any region, so it cannot be told whether this product keeps what the activity requires.

In human cells GEMIN5 binding to the SMN 3'-UTR increases SMN translation, and GEMIN5 bound to ribosomes through its N-terminal domain dampens global protein synthesis, so it regulates translation in both directions. The horse product keeps the relevant RNA-binding regions and is in the same PANTHER subfamily, which justifies transfer as an inference. The full-length product F6WDM0 already carries regulation of translation in GOA (Ensembl Compara IEA), and the gene review of this product uses the same term in its core functions, so the prediction is correct but not novel.

MAP2K2 (MEK2) is a dual-specificity MAP kinase kinase that itself phosphorylates the threonine and tyrosine of the ERK1/ERK2 activation loop, so the gene product performs protein phosphorylation rather than merely being required for it. Human MEK2 phosphorylates and activates ERK1 in vitro (PMID:8388392), and horse MAP2K2 descends from PAINT node PTN000684494, which asserts MAP kinase kinase activity with human MEK2 among its seeds. A0A9L0SHX8 is an alternative horse MAP2K2 gene model that retains human residues 1-363 with Lys101, Asp194 and Ser222/Ser226 conserved, but carries an unrelated C-terminal tail; the prediction is judged as a claim about the horse MAP2K2 gene product, and the catalytic competence of this particular model is untested. Neither this entry nor any other horse MAP2K2 product (A0A9L0SF07, F6YUU0, A0A9L0R4J0, A0A9L0SD26) carries GO:0006468 or a descendant in GOA; MAPK cascade and ERK1 and ERK2 cascade are signal-transduction terms, not is_a descendants of protein phosphorylation, and the kinase MF terms are in a different aspect. The broad BP is therefore correct and novel. Training-set membership is unknown.

Human MTMR9 raises the catalytic activity of its active partners MTMR6 and MTMR8 (direct assays, PMID:22647598), and mouse MTMR9 binds MTMR7 through its coiled coil and increases MTMR7 activity (PMID:12890864). The human and mouse gene reviews accept this term, and the horse protein is a 1:1 ortholog in the same PANTHER subfamily (PTHR10807:SF56), 97.8% identical over 507 aligned residues and retaining the C-terminal partner-binding region. Horse GOA lacks this process term (it carries only enzyme regulator activity), so this is a correct novel prediction by justified orthology transfer. It describes regulation of partner catalysis, not intrinsic phosphatase activity; the 42-residue deletion at the GRAM/phosphatase-like domain junction leaves the activity of this exact product untested.

Human MTMR9 contributes to negative regulation of autophagy as the catalytically inactive activating subunit of the MTMR8-MTMR9 PtdIns(3)P phosphatase complex (PMID:22647598); the human and mouse reviews keep this term as non-core, and PAINT places it on node PTN002246859 within the MTMR9 lineage, which the horse gene descends from. The OpenScientist audit found that the horse sequence retains the inactive PTP-like domain and the C-terminal coiled-coil dimerization region, supporting transfer of the partner-dependent mechanism despite the 42-residue deletion. Horse GOA lacks this term, so it is a correct novel prediction. It is a context-dependent, non-primary effect: MTMR9 knockdown does not enhance autophagic clearance of TDP-43 in neurons.

Formation of the MTMR6-MTMR9 heteromer increases protein levels of both partners, attributed to inhibition of their degradation (PMID:19038970), and MTMR9 binding raises MTMR8 stability. The human and mouse MTMR9 reviews accept protein stabilization as part of a core partner-binding function. The horse product is a 1:1 SF56 ortholog that retains the C-terminal coiled coil through which MTMR9 binds its partners, so the transfer is justified on the same grounds as the regulatory terms; it is an inference rather than an observation on the horse protein. Horse GOA carries no stabilization term, so the prediction is correct and novel, with the caveat that the 42-residue internal deletion has not been tested for effects on folding or partner stabilization.

Human MTMR9 forms heterodimers with the active myotubularins MTMR6, MTMR7 and MTMR8 (PMID:22647598, PMID:19038970), and mouse MTMR9 forms a complex with MTMR7 through its coiled coil (PMID:12890864). The human review accepts this IDA term as the most precise complex term GO offers. The horse product retains the coiled coil and is a same-subfamily ortholog, so membership in a protein-containing complex transfers. Horse GOA has no complex annotation, making this a correct novel prediction, although the term is broad.

MTMR9 binds the myotubularin lipid phosphatases MTMR6, MTMR7 and MTMR8, and that binding transfers to the horse ortholog. The predicted term, however, names protein phosphatase binding. The human and mouse MTMR9 reviews replace this term with phosphatase binding (GO:0019902) because the partners dephosphorylate phosphoinositides rather than proteins. No evidence identifies a protein-dephosphorylating partner of MTMR9, and the binding data do not refute the term outright given the ontology-scope question, so the exact claim remains uncertain.

Enzyme regulator activity is supported by human MTMR9 stimulation of MTMR6 and MTMR8 (PMID:22647598) and mouse MTMR9 stimulation of MTMR7 (PMID:12890864), transferred to the 1:1 horse ortholog that is 97.8% identical over its length. The exact term is already in horse GOA for this product, so the prediction is correct but not novel. The horse gene review refines it to phosphatase activator activity, which this broad prediction entails.

Complex formation with MTMR9 increases the activity of MTMR6 and MTMR8 and shifts their phosphoinositide substrate preference (PMID:22647598); the human and mouse MTMR9 reviews accept this term as core. MTMR9 does this work directly as the activating subunit of the heterodimer, without catalysing the dephosphorylation itself. The horse product is a same-subfamily ortholog retaining the partner-binding coiled coil, and horse GOA has no corresponding process term, so this is a correct novel prediction by orthology transfer, with the internal-deletion caveat.

A0A9L0SKW1 lacks the N-terminal signal peptide: its N-terminus MDSQ-VFGD replaces the hydrophobic leader that aligns with human signal peptide 1-27 (maximum Kyte-Doolittle hydropathy 0.25 versus 2.93 for human and 3.13 for the leader-bearing horse isoform X1), and the record carries no SIGNAL or transmembrane feature; its Secreted annotation comes from an ARBA family rule. The product therefore lacks the route by which photomedin-1 reaches the matrix. That does not refute the process term, because the mechanism by which OLFML2A takes part in extracellular matrix organization has not been established: mouse photomedin-1 is a matrix-associated glycosaminoglycan-binding protein, and the Manabe et al. ECM screen behind the mouse acts_upstream_of_or_within annotation assayed several properties (ECM assembly, interactions with ECM molecules, glycosaminoglycan modification, cell adhesion) without the cached abstract saying which one Olfml2a satisfied. With the participation mechanism unknown, a contribution by the leaderless product through some other route is neither shown nor excluded, and the gene-level support for the term is itself weak, so the prediction can be neither confirmed nor refuted for this accession.

Horse GOA carries no molecular-function annotation, so the term would be novel for the gene. The experimental basis is mouse photomedin-1, secreted as a disulfide-bonded dimer, and curated human UniProt records OLFML2A as a homodimer by similarity. That disulfide-linked dimer forms in the oxidizing secretory pathway, which A0A9L0SKW1 cannot enter because it lacks the N-terminal signal peptide (its N-terminus MDSQ-VFGD replaces the leader matching human 1-27). The accession also lacks the 22-residue segment opposite human 123-144, but retains the region matching the human coiled coil (157-183) and the olfactomedin domain. Which region mediates photomedin-1 self-association, and whether a cytosolic leaderless product could dimerize non-covalently, is unresolved, so the prediction can be neither confirmed nor refuted for this product.

The term is true of the OLFML2A gene: mouse photomedin-1, the characterized member of PANTHER subfamily SF29, was identified in a screen for novel ECM proteins, is secreted, and binds chondroitin sulfate E and heparin, and the horse locus encodes a leader-bearing product. It is not correct for the sequence ProtNLM2 was given. A0A9L0SKW1 lacks the N-terminal signal peptide: its N-terminus MDSQ-VFGD replaces the hydrophobic leader that aligns with human signal peptide 1-27 (maximum hydropathy 0.25 versus 2.93 for human and 3.13 for the leader-bearing horse isoform X1), and the record carries no SIGNAL or transmembrane feature; its Secreted annotation comes from an ARBA family rule. Without a signal peptide this product has no route into the secretory pathway and cannot be deposited in extracellular matrix. Novelty against horse GOA (extracellular region only) is not at issue; the prediction fails on product architecture.

Mouse photomedin-1 bound chondroitin sulfate E and heparin preferentially among a panel of ECM components, and the horse gene review assigns this activity to the gene by sequence similarity. The olfactomedin domain is intact in A0A9L0SKW1, but the binding partners are extracellular glycosaminoglycans that the protein can only reach after secretion. A0A9L0SKW1 lacks the N-terminal signal peptide: its N-terminus MDSQ-VFGD replaces the hydrophobic leader that aligns with human signal peptide 1-27 (maximum hydropathy 0.25 versus 2.93 for human and 3.13 for the leader-bearing horse isoform X1), and the record carries no SIGNAL or transmembrane feature; its Secreted annotation comes from an ARBA family rule. The leaderless product therefore cannot carry out extracellular matrix binding; the activity belongs to the leader-bearing product of the locus.

Human-cell experiments establish that OMA1 is the zinc metalloprotease that cleaves OPA1 (PMID:20038677) and DELE1, and the horse sequence retains the HExxH catalytic motif and third zinc ligand at 66.47% overall paired identity, so the proteolysis claim is a justified transfer and is biologically correct. It is not novel, since GO:0006508 is already in horse GOA (InterPro2GO). It is, however, materially less specific than the gene's accepted core process GO:0034982 mitochondrial protein processing (also in horse GOA), which is_a protein processing is_a proteolysis (OLS4 ancestry). The gene review accordingly replaces GO:0006508 with GO:0034982. Training-set membership is unknown.

The horse gene model carries the complete human PEA15 core at 100% identity (with a 46-residue N-terminal extension) in PANTHER subfamily PTHR48169:SF1, so human PEA15 evidence transfers to it. That evidence shows PEA15 regulates glucose transport rather than performing it. Overexpressed human PEA15 in L6 myotubes raises surface GLUT1 and inhibits insulin-stimulated glucose transport and GLUT4 recruitment, and this effect is mediated indirectly through PEA15 binding to the phospholipase D1 D4 domain and the consequent increase in PKC-alpha activity. The human gene review keeps only the regulatory term GO:0046325 (negative regulation of D-glucose import across plasma membrane), as non-core, and notes that PEA15 has no transporter activity; the horse GOA set has no carbohydrate-transport or glucose-import term. GO:0008643 asserts participation in directed carbohydrate movement, and a regulation term is not an is_a descendant of the regulated process, so the existing regulatory evidence neither supports nor overlaps the prediction. PEA15 is not shown to take part in the transport step itself, but no evidence directly excludes a participatory role, so the prediction remains uncertain rather than refuted.

The only experimental link between PPP4R4 and this process is a pooled RNAi screen in mouse zygotes, in which knockdown of the ortholog Ppp4r4 (same PANTHER subfamily PTHR21467:SF0) was one of 35 knockdowns scored as defective in a combined hatching and outgrowth assay. That is a single unreplicated screen result, it measures necessity rather than participation, and the authors themselves flag that Ppp4r4 is not expressed at the stages where it appears to be required. No mechanism connects the cytosolic PP4c-PP4R4 complex, whose substrates are unknown, to zona escape. PubMed searches for PPP4R4 or KIAA1622 with embryo or blastocyst found no other study. Neither horse product carries the term in GOA, and the horse gene review does not address it. The prediction is neither supported nor refuted; a validated knockout or rescue phenotype with a traced mechanism would be needed.

Human PP4R4 is an experimentally isolated stable cytosolic partner of PP4c, and the PP4c-PP4R4 complex has lower phosphatase activity than free PP4c, which establishes a noncatalytic regulatory role. Horse PPP4R4 is in the same PANTHER subfamily (PTHR21467:SF0) and PAINT node, and A0A9L0S961 keeps the HEAT repeats and the PP4c-binding residues, so the transfer is justified. The term is the core molecular function in the horse gene review. It is not novel for the gene: the canonical horse product A0A9L0S938, which is identical to A0A9L0S961 except at the extreme C-terminus, already carries GO:0019888 by IBA.

Human PP4R4 binds PP4c specifically, and not PP2Ac or PP6c, in a stable cytosolic complex, so PPP4R4 is a subunit of a serine/threonine phosphatase complex; the transfer to horse rests on shared subfamily PTHR21467:SF0 and conserved PP4c-binding residues. The exact term is already carried by IBA on the canonical horse product A0A9L0S938 of the same gene, so the prediction is not novel. The horse gene review refines this complex to the more specific child GO:0030289 protein phosphatase 4 complex, but that refinement is a proposal rather than an existing annotation, so the prediction matches the existing term rather than being less precise than one.

GO:0006470 asserts that PTPRN2 itself takes part in removing phosphate from proteins. PTPRN2 (IA-2beta/phogrin) is an IA-2 family pseudophosphatase (PANTHER PTHR46106:SF5) with no detectable activity on phosphoprotein or pNPP substrates. Its only demonstrated intrinsic catalytic activity, shown for the rat ortholog in the same subfamily, is dephosphorylation of phosphoinositide lipids. That is a different substrate class and supports phosphatidylinositol phosphate phosphatase activity, not protein dephosphorylation. The horse and human gene reviews remove the protein tyrosine phosphatase activity rows and the human GO:0006470 row on these grounds. Reported protein-directed effects are indirect. Phogrin protects PTP1B from oxidation and so allows PTP1B to dephosphorylate the insulin receptor, and a correlative cancer study proposes that PTPRN2 promotes STAT1 dephosphorylation without identifying an enzyme. At most these would support regulation of protein dephosphorylation, in which another phosphatase does the work. The horse model also lacks the conserved C-terminal end of the PTP-like domain. The prediction transfers a PTP-fold process to a pseudophosphatase and is refuted.

The horse SHLD2 gene carries this term on its canonical product A0A9L0TUF0; whether this product carries it is unresolved. Human SHLD2 promotes NHEJ as part of shieldin: its N-terminus binds SHLD3-REV7 for recruitment, and its C-terminal OB-fold module (SHLD2C) binds single-stranded DNA with SHLD1 to protect ends. A0A9L0RGD6 keeps the recruitment segment, OB folds A and B and the four zinc-ribbon cysteines of OB fold C, but conservation ends after horse residue ~811: its Pfam SHLD2 C-terminal match spans 733-804 (72 residues) against 772-935 (164 residues) in the canonical model, and it lacks the equivalent of human iso-1 731-835, the distal half of the SHLD1-interacting region (652-822). Human OB-fold point mutant m1 and the SHLD2S variant disrupting OB fold B still reach DSBs and assemble shieldin but fail to suppress HR, so the module is required. Neither mutation corresponds to this C-terminal truncation, and no experiment has tested whether a SHLD2 that keeps OB folds A, B and the zinc ribbon binds ssDNA or SHLD1. The missing segment is part of the required module, but its loss is not shown to abolish the function, and the retained recruitment and partial SHLD1-binding regions could still contribute. Neither positive refutation nor a fully accounted mechanism applies, so the call is uncertain.

HPA reports an additional actin-filament pool of human SHLD2, but the human SHLD2 review accepts only the main nucleoplasm location and no GO annotation or experiment supports actin association. The secondary signal is a single antibody-based observation with no known targeting mechanism, and family membership does not establish that the horse protein shares it. No PubMed literature links SHLD2 to actin. The term is absent from the horse gene's annotations and cannot be validated or refuted.

Human SHLD2 is recruited to DSBs via the 53BP1-RIF1 axis through its N-terminal SHLD3-REV7-binding segment, and human OB-fold mutants still localize to DSB sites, so recruitment does not depend on the C-terminal module this model has lost. A0A9L0RGD6 is in the single SHLD2 subfamily and keeps the N-terminal recruitment region, so the location transfers by orthology; this is an inference, not a horse observation. The horse SHLD2 gene already carries site of double-strand break (IBA and IEA) on its canonical product A0A9L0TUF0, so the prediction is correct but not novel.

Human SHLD2 is mainly nucleoplasmic by HPA immunofluorescence and relocalizes into nuclear damage foci. A0A9L0RGD6 is an orthologous SHLD2 model that keeps the N-terminal recruitment region, so the nuclear location transfers by orthology; this is an inference for horse. The horse SHLD2 gene already carries nucleoplasm (IEA ortholog transfer) on its canonical product A0A9L0TUF0, so the prediction is correct but not novel.

The horse SHLD2 gene carries this term on its canonical product A0A9L0TUF0; whether this product carries it is unresolved. Human SHLD2 suppresses HR by blocking end resection through ssDNA binding by the C-terminal SHLD2-SHLD1 OB-fold module (SHLD2C). OB-fold mutant m1 and the SHLD2S variant disrupting OB fold B localize to DSBs and assemble shieldin yet fail to suppress HR, so recruitment alone is not enough. A0A9L0RGD6 keeps the recruitment segment, OB folds A and B and the OB-fold-C zinc ribbon; its Pfam SHLD2 C-terminal match spans only 733-804 (72 residues) against 772-935 (164 residues) in the canonical model, and it lacks the distal half of the human SHLD1-interacting region, which is replaced by an unrelated Arg/Gly/Pro-rich tail. The tested human mutations differ from this C-terminal truncation, and whether the partial module still binds ssDNA or SHLD1 has not been measured. The anti-HR mechanism therefore depends on a module this product keeps only in part, and loss of the function is neither shown nor excluded, so the call is uncertain.

Shieldin loss impairs class-switch recombination in mammalian B cells, and the horse SHLD2 gene carries this term (IEA ortholog transfer) on its canonical product A0A9L0TUF0. SHLD2 contributes to CSR through NHEJ-promoting end protection by shieldin, which uses the N-terminal SHLD3-REV7 recruitment segment and the C-terminal SHLD2-SHLD1 OB-fold ssDNA-binding module. A0A9L0RGD6 keeps the recruitment segment, OB folds A and B and the OB-fold-C zinc ribbon, but has a truncated OB fold C (Pfam SHLD2 C-terminal match 733-804, 72 residues, against 772-935, 164 residues, in the canonical model) and lacks the distal half of the SHLD1-interacting region. No experiment shows that this truncation abolishes end protection or CSR, and no horse B-cell or CSR evidence exists. The missing segment does not fully account for the gene's role in CSR, and no evidence refutes the term for this product, so the call is uncertain.

Metal ion binding is true of the horse SIRT5 gene: sirtuins bind one structural zinc ion through a four-cysteine zinc ribbon, human SIRT5 is curated as binding one Zn(2+) per subunit (PMID:22076378), and the full-length horse product A0A9L0T9B1 retains all four zinc-ligand cysteines and carries GO:0008270 zinc ion binding. This accession, however, lacks the feature the activity depends on. F6S899 is a defective gene model whose transcript skips a 124-nt coding exon, so its C-terminal third is translated out of frame; mapping the human zinc ligands onto it shows Cys166 and Cys169 retained but Trp at the positions of Cys207 and Cys212. With half of the tetrathiolate zinc site missing, this exact sequence has no established metal-binding site, and no other metal site is known for the family. The prediction is therefore incorrect for the product scored.

TRAF2 knockdown suppresses TCR-dependent interleukin-2 production in the primary mammalian study. The selected horse protein retains the full human adaptor architecture at 92.0% paired identity, supporting conserved positive control of T-cell cytokine production. This inference does not require TRAF2 itself to catalyze ubiquitin transfer. The supported annotation is absent from the frozen horse GOA; model training membership is unknown.

Recombinant human TRAF2 directly binds the CD40 cytoplasmic domain. Full coverage of the characterized human sequence, including the C-terminal receptor-binding TRAF region, supports the same molecular interaction for horse F7BIV4. The supported annotation is absent from the frozen horse GOA; model training membership is unknown.

A reconstituted TRAF2–cIAP1 complex establishes recruitment of a genuine E3 ligase by TRAF2. The conserved horse scaffold supports membership in a ubiquitin ligase complex; this annotation does not assert autonomous catalytic activity of TRAF2. The supported annotation is absent from the frozen horse GOA; model training membership is unknown.

TRAF2 is the receptor-associated adaptor that recruits cIAP1/2 into TNF receptor superfamily signaling assemblies. The conserved horse architecture supports complex membership independently of the disputed intrinsic E3 mechanism. The supported annotation is absent from the frozen horse GOA; model training membership is unknown.

TRAF2-deficient mouse B cell lines are defective in CD40-mediated, TNF-dependent IgM production, and TRAF2-deficient mice have defective T helper-dependent antibody responses; mouse Traf2 carries this term by IMP. Horse F7BIV4 is a full-length member of the same TRAF2 orthologous subfamily with the receptor-binding MATH domain conserved, which justifies transfer. The horse gene has no immunoglobulin-production annotation in GOA. The role is a downstream, non-core consequence of TRAF2's CD40 adaptor function.

The predicted term matches an existing electronic GOA annotation for this horse gene (UniRule-based IEA on this product and on F7BV54), but that overlap is not biological support, and the gene review marks the term as over-annotated. Scored for this exact product, the prediction rests on deubiquitinase activity, and whether this splice form has it is unresolved. A0A9L0T7K6 uses an alternative 3' terminal exon. It keeps the MIT and rhodanese-like regions and the catalytic Cys785 and His1066, but only 328 of the 333 USP-domain residues align, and the conserved C-terminal USP segment (human ...AAYILFYTSLGPRVTDVAT) is replaced by an unrelated tail (...AASFSLHFREISQS). No assay of this form exists, so it is unclear whether the truncated USP fold is catalytically competent. At the gene level the term is also uncertain. USP8 is a cysteine-type deubiquitinase whose effects on protein turnover depend on the substrate. It stabilizes Nrdp1 and STAM by removing ubiquitin, yet its knockdown inhibits degradation of stimulated EGFR and Met. Human curators captured these roles with regulatory terms (regulation of protein stability, negative regulation of lysosomal protein catabolic process) rather than GO:0006511, and no experimental annotation or PAINT node assigns GO:0006511 to USP8.

WDPCP is a CPLANE subunit required for cilium formation: mouse CPLANE recruits peripheral IFT-A to the basal body, and a mouse Wdpcp in-frame deletion abolishes ciliogenesis. Horse A0A3Q2KRK8 is in the same PANTHER subfamily (SF5) as these orthologs and descends from PAINT node PTN001007408, which carries cilium organization; the canonical horse WDPCP product A0A9L0T876 carries GO:0044782 cilium organization by IBA, accepted in its gene review. GO:0044782 is_a GO:0120036 is_a GO:0030030 (OLS4), so the prediction is correct but less specific than an annotation the gene already has. The OpenScientist synthesis reaches the same conclusion on term breadth. The actin protrusion and migration phenotypes are real but do not by themselves warrant a more specific process term here.

Mouse Wdpcp has a cytoplasmic pool on actin filaments and focal adhesions, and Xenopus Fritz concentrates at the cell cortex; both orthologs share PANTHER subfamily SF5 with horse A0A3Q2KRK8, which retains the full propeller and solenoid. The prediction is therefore correct. This product already carries GO:0005930 axoneme (IEA, accepted in the gene review on direct mouse and Xenopus axonemal localization), and axoneme is a descendant of cytoplasm via ciliary plasm (OLS4); the cortical pool is likewise better captured by cell cortex, the replacement the gene review proposes for plasma membrane. Cytoplasm is thus a less precise restatement of supported locations.

Endogenous mouse Wdpcp is detected along the ciliary axoneme and in a ring at the ciliary base, and Xenopus Fritz localizes along axonemes and near basal bodies; horse A0A3Q2KRK8 is in the same PANTHER subfamily. Cilium is therefore correct. The gene already carries axoneme on this product and axonemal basal plate (IBA) on the canonical product A0A9L0T876, both accepted in the gene reviews, and both are descendants of GO:0005929 cilium (OLS4). The prediction is correct but less specific than existing supported annotations.

Mouse Wdpcp localizes to the actin cytoskeleton and is required to recruit Sept2 to actin filaments, and in Xenopus it acts through septins and actomyosin at the cell cortex; this orthology-based transfer within PANTHER subfamily SF5 supports the term. GO:0005930 axoneme, already carried by this product and accepted in the gene review, is a descendant of GO:0005856 cytoskeleton (OLS4), so the prediction is less specific than an existing supported annotation. The actin-associated pool has no more specific annotation in GOA, but the term as predicted remains an ancestor of the accepted axoneme location.

WEE1 is a conserved CDK-inhibitory kinase, but mammalian oocyte studies distinguish somatic WEE1 from WEE2/Wee1B. A mitotic checkpoint mechanism does not automatically establish regulation of meiosis I. Wee1B/WEE2 is the oocyte-specific MPF-inhibitory kinase that maintains prophase I arrest, but immunodetected wee1 is present in mouse oocyte nuclei during acquisition of meiotic competence (antibody specificity versus Wee1B not established), and Wee2 knockout mice retain fertility, so a somatic WEE1 contribution to meiosis I is neither shown nor excluded. No perturbation of somatic WEE1 in mammalian meiosis I was found in a PubMed search.

Human experiments place the soluble checkpoint kinase in the nucleus. Full-length horse sequence conservation (same PANTHER subfamily SF72 as human WEE1, which carries nucleoplasm from Reactome CDK1-phosphorylation reactions) supports nucleoplasmic activity. The horse gene carries only nucleus and nucleolus, so nucleoplasm is a novel, more specific location.

The characterized WEE1 checkpoint pool is nuclear, with additional cytoplasmic annotations. A PubMed search found membrane association for the paralog PKMYT1/Myt1 and cortical-node localization of fission-yeast Wee1, but no evidence establishing plasma-membrane recruitment of vertebrate WEE1. A soluble kinase can associate peripherally with a membrane, so absence of a transmembrane helix does not refute the prediction.

The prediction assigns a WEE2/Wee1B oocyte process to the somatic WEE1 paralog. The selected horse F6TY09 sequence is a full-length WEE1 clade member, not the horse WEE2 protein, and the focused WEE1/WEE2 comparison places F6TY09 at ~90-95% identity to human and mouse WEE1 but only ~50% identity to WEE2 orthologs. The mouse experiment that supports female pronucleus assembly specifically showed Wee1B depletion blocking pronucleus formation while Wee1A-depleted oocytes activated correctly. Horse has a separate WEE2 entry, A0A9L0SIB9, and the WEE2 clade, not F6TY09, carries GO:0035038 across mammalian annotations. Later Wee2 knockout and possible-compensation evidence cautions against claiming that WEE2 is absolutely indispensable, but does not support transferring this WEE2-clade process back to somatic WEE1.

Protein phosphorylation follows directly from the conserved WEE1 tyrosine-kinase reaction. Mouse Wee1 has direct-assay evidence for protein phosphorylation, and the horse gene carries no phosphorylation process term, so the prediction is novel for this gene. It is correct but broad relative to the specific peptidyl-tyrosine phosphorylation of CDK1 Tyr15.

WEE1 catalyzes inhibitory CDK phosphorylation, which supports protein phosphorylation. A separate positive-regulation-of-phosphorylation process requires identifying the regulated reaction; the direct catalytic reaction alone does not establish that regulatory relationship, and effects on downstream CDK phosphorylation are inhibitory.

Mammalian studies identify WEE2/Wee1B as the oocyte-enriched paralog, whereas this horse sequence is WEE1. The broad female-meiosis term is not validated by conserved kinase activity alone. Redundancy and species-specific expression leave possible participation unresolved rather than proving absence. Immunodetected wee1 is present in mouse oocytes, and Wee2 knockout mice retain fertility, so neither indispensability of WEE2 nor absence of a WEE1 contribution is established.

Human ZDHHC23-dependent KCNMA1 palmitoylation is experimentally supported (PMID:22399288). The horse sequence retains the DHHC catalytic cysteine (human289/horse283), with 91.07% identity over 98.53% of the human sequence. Transfer of protein-cysteine S-palmitoyltransferase activity is justified and agrees with the exact horse IBA/IEA term. This judgment rests on substrate experiments and sequence conservation, not ARBA agreement. Training-set membership is unknown.

The target is the TTC39C ortholog (PANTHER subfamily SF1, the same subfamily as zebrafish ttc39c), and the term is absent from its GOA annotations, so the prediction would be novel. The only experimental support is the zebrafish morpholino screen of Xu et al. 2015 (PMID:25860617), in which more than 64% of ttc39c morphants had misoriented heart tubes; the authors attribute this, like the otolith and body-curvature defects, to reduced number and length of Kupffer's vesicle cilia. Motile cilia of the left-right organiser do participate in laterality determination, so the prediction is biologically plausible, but the evidence is a single knockdown study with no mammalian loss-of-function data, and the PANTHER PAINT assertions for PTHR31859 propagate only cilium assembly and otolith morphogenesis, not a laterality term. The merged gene review accepts cilium assembly as the supported process and removes the parallel organismal readout (otolith morphogenesis) as a downstream ciliopathy phenotype rather than process participation; the heart asymmetry phenotype is of the same kind. No PubMed record links TTC39C to laterality outside this screen. The evidence therefore neither establishes nor refutes heart left/right asymmetry determination for macaque TTC39C.

GOA has no annotations for this accession, so the term is novel. The protein has a Pfam RVT_1 reverse-transcriptase domain (174-349) with an intact YLDD motif C and aligns continuously with the Dictyostelium DIRS-1 ORF2 RT-RNase H in the same PANTHER subfamily (PTHR33050:SF7), whose RT is shown genetically to supply cDNA synthesis for DIRS-1 retrotransposition. RNA-directed DNA polymerase activity extends a DNA primer strand on an RNA template, so DNA engagement is an inherent part of the inferred catalysis. GO:0003677 is not an ancestor of GO:0003964 RNA-directed DNA polymerase activity (OLS4), so LSP does not apply, although DNA binding is a minimally informative description of an enzyme whose DNA contact is with its own substrate. The support is a family transfer with an intact motif, not an assay of the mussel protein.

GOA has no annotations for this accession. The recombination step of DIRS-type elements is the site-specific recombination by which the element-encoded tyrosine recombinase (YR) integrates the circular cDNA intermediate. In DIRS-1 the YR is a separate ORF from the RT-RNase H protein, and A0A8B6BFL6 has no recombinase domain and no detectable similarity to the DIRS-1 YR (z = -0.7 against shuffled controls), so it cannot perform the strand-exchange chemistry. That does not refute participation in the process: the protein makes the cDNA that is recombined, the enzyme that converts the single-stranded circle to the double-stranded recombination substrate has not been identified, and a scaffolding or binding contribution to a YR-containing complex has neither been shown nor excluded. With no positive refutation and no evidence of participation, the prediction is uncertain. The gene review assigns this protein to reverse transcription within retrotransposition and does not annotate DNA recombination.

GOA has no annotations for this accession. DIRS-type elements are defined by a tyrosine recombinase in place of an integrase or endonuclease, and that recombinase mediates genomic integration of the element. A0A8B6BFL6 is the RT-RNase H protein: it has no integrase or recombinase domain and does not align with the DIRS-1 tyrosine recombinase, so it does not catalyse the joining step. The RT generates the cDNA that is integrated, which establishes necessity rather than participation. An accessory role, such as completing the double-stranded circular substrate or acting in a pre-integration complex with the recombinase, has not been tested in any DIRS element and is not excluded by the sequence. The prediction therefore rests on missing evidence rather than refutation and is uncertain. The gene review assigns this protein to retrotransposition through reverse transcription and does not annotate DNA integration.

The prediction is an intrinsic lipid 3-phosphatase activity, which is not among the gene's GOA annotations. A0A8B6GS20 is a partial MTMR9-type gene model whose myotubularin-like region (residues 362-607) aligns to the C-terminal part of human MTMR9 and is placed in PANTHER PTHR10807:SF73 with C. elegans MTM-9, which is described experimentally as phosphatase-inactive. The retained CX5R-equivalent P-loop reads VHGSEGFDTT, with Gly at the catalytic Cys position and Thr at the Arg position, the same kind of substitution seen in human MTMR9 and worm MTM-9. PAINT records the loss of GO:0004438 (and GO:0106018) by IRD NOT rows at the MTMR9 clade node PTN000086907, and the gene review treats the protein as a noncatalytic regulatory subunit that binds and activates an active myotubularin partner. MTMR9-family proteins raise partner 3-phosphatase output without catalysis themselves (PMID:22647598), so participation in phosphatidylinositol dephosphorylation is supported while the predicted intrinsic activity is not. This is a catalytic activity transferred to a fold-retaining pseudophosphatase.

The RPC4-specific domain identifies the C53 subunit of nuclear RNA polymerase III, and reconstitution of the characterized yeast polymerase shows that the C37-C53 module is needed for terminator recognition and reinitiation. The predicted term is correct and is already carried by NCU01245 in GOA (InterPro2GO IEA), where the gene review accepts it as appropriately general because polymerase III transcribes 5S rRNA, U6 snRNA and other RNAs besides tRNAs; the narrower tRNA transcription by RNA polymerase III (GO:0042797, IBA) is also present. The role is participation as a noncatalytic enzyme subunit.

The SPX-permease architecture and PANTHER Pho91 subfamily (PTHR10283:SF92) placement support phosphate transport by transfer from the characterized fungal Pho87/Pho90/Pho91 transporters; yeast Pho91 is a vacuolar phosphate transporter that exports phosphate to the cytosol (PMID:17804816, PMID:41324358). The gene already carries the same generic term from an ARBA rule, which the gene review refines to the more specific, IBA-supported phosphate transmembrane transporter activity (GO:0005315, a descendant of the predicted term). The prediction is therefore correct but less precise than an existing supported annotation. The unresolved plasma versus vacuolar location does not undermine generic transmembrane transport.

NCU02539 carries the InterPro MCM_4 signature and is classified in the Mcm4 subfamily PTHR11630:SF66, the subfamily whose PAINT node gives it mitotic DNA replication initiation and DNA strand elongation. Budding-yeast experiments show that Mcm2-7 is loaded at origins during licensing and is needed for fork progression after initiation, which together make up nuclear DNA replication. GO:0033260 is already in GOA for this gene and the gene review accepts it as the level that covers both mitotic and premeiotic replication, so the prediction is correct but not novel.

Mcm4-subfamily membership places NCU02539 in the Mcm2-7 helicase. Mcm2-7 is loaded onto origins in the pre-replicative complex and becomes the motor of the CMG helicase that unwinds origin DNA, which is the Mcm2-7 contribution to initiation. GO:0006270 is already in GOA for this gene and is accepted in the gene review as a core process at the level that spans mitotic and premeiotic S phase, alongside the more specific IBA term GO:1902975. The prediction is correct but not novel. It concerns the Mcm4 subunit as part of the complex, not autonomous initiation activity.

Purified yeast Mcm2-7 and Mcm4/6/7 complexes bind single-stranded DNA in an ATP-dependent way, and mutational analysis implicates Mcm4 in the single-stranded DNA binding of the complex. GO:0003697 single-stranded DNA binding, an is_a descendant of GO:0003677, is already annotated to this gene by PAINT (MCM2-7 node) and is accepted in the gene review, which refines the generic DNA-binding row to it. The generic prediction is correct but loses the substrate specificity, so it is less precise. The binding is a property of the assembled ring, not of an independent DNA-binding subunit.

NCU03033 carries the NF-YA/Hap2 signature (InterPro IPR001289) and the curated RuleBase family assignment, and the family's characterised fungal and human members are interchangeable CCAAT-box-binding transcription factor subunits that act on RNA polymerase II promoters. This is a justified family transfer, not an observation on the N. crassa protein itself. GOA already carries the exact predicted term (IEA), which the gene review modifies to the more specific GO:0000981 DNA-binding transcription factor activity, RNA polymerase II-specific (is_a descendant of GO:0003700), inherited by IBA from PAINT node PTN000290575 and accepted as the core function. The prediction is correct but less precise than that supported annotation.

Nucleus is the site of action of the CCAAT-binding factor, which binds CCAAT boxes in target promoters; this follows from the curated NF-YA/Hap2 family assignment of NCU03033 and its membership in the CCAAT-binding factor complex (IBA, PTN000290576). In N. crassa itself a heteromeric CCAAT-binding activity was purified from nuclear protein, although that study did not identify NCU03033 as one of its subunits. GOA already carries the exact term (IEA), which the gene review accepts, so the prediction is correct but not novel.

The target is a Ubc9-family SUMO E2, supported by modifier-specific CDD UBCc_UBE2I and FunFam Ubc9 assignments, full-length similarity to characterized Ubc9, and the curated SUMO-specific PAINT inference. Characterized Ubc9 homologs form thioester intermediates with SUMO. Transferase activity is therefore correct but less precise than the existing SUMO conjugating enzyme activity. This broad GO result does not validate the separate ubiquitin-specific paragraph.

Cytoplasm is correct and already annotated to the gene, which also carries the more specific actin cortical patch localization. PANTHER places the target in the Rvs167 subfamily (PTHR47174:SF1), and its BAR-plus-SH3 architecture matches the characterized yeast endocytic adaptor Rvs167, which localizes to cortical actin patches at endocytic sites on the cytoplasmic face of the plasma membrane. The gene already carries cytoplasm (InterPro2GO IEA) and the more specific actin cortical patch (IBA, accepted in the gene review), which is a part_of descendant of cytoplasm via cell cortex; the gene review refines the cytoplasm row to that term. The ProtNLM2 string match points to medial cortex, a fission-yeast-specific term the gene review marks over-annotated, but the emitted cytoplasm term does not depend on it.

NCU04937 is an uncharacterised 114-residue orphan ORF in which a single run of 50 glutamines (residues 55-104) makes up about half the sequence, under a predicted coiled coil. It has no conserved domain; its only family signature, a weak PANTHER PTHR13270:SF14 match, lies over the polyglutamine run and reflects low-complexity sequence rather than a shared domain. The prediction cites human THAP11 (Q96EK4; phmmer score 29.1), whose DNA binding resides in its THAP zinc-finger domain, which NCU04937 lacks; shared glutamine-rich content does not transfer that function. The gene has no GOA annotations, no ortholog was detected in other Neurospora species, and a PubMed search for NCU04937 returned no publications. Nothing supports or refutes DNA binding, so the prediction is uncertain.

The glycerol-kinase-specific family assignment and classical N. crassa glycerol-dissimilation enzymology ground transfer of glycerol phosphorylation. This directly produces glycerol-3-phosphate, supporting the existing biosynthetic-process annotation. The predicted metabolic process is biologically correct but less precise than biosynthesis.

NCU07379 has a bZIP superfamily domain and is placed in the FlbB clade of the fungal AP-1-like subfamily PTHR40621:SF6. Aspergillus nidulans FlbB, the characterised member of that clade, is a bZIP factor that activates transcription of flbD and, with FlbD, of brlA. The prediction is therefore correct. The gene already carries GO:0003700 by InterPro mapping and the more specific descendant GO:0001228 (RNA polymerase II-specific activator activity) by PAINT, and the gene review accepts GO:0001228 as the core function, so the generic transcription-factor term is less precise than an existing supported annotation and is not novel. The N. crassa target genes are unknown.

Correct but less precise than an existing supported annotation. NCU08595 is in the Rpf2 subfamily (InterPro IPR039770; PANTHER PTHR12728:SF0), and characterized fungal Rpf2-Rrs1 complexes bind 5S rRNA, recruit the 5S RNP into pre-60S particles and are required for 27SB pre-rRNA processing, so participation in LSU-rRNA maturation transfers by subfamily membership. The predicted term is already in GOA for this gene as an InterPro2GO IEA, and the gene carries its is_a descendant GO:0000463 (maturation of LSU-rRNA from tricistronic rRNA transcript) by IBA from the PAINT Rpf2 node. The gene review refines GO:0000470 to GO:0000463, so the prediction is the general form of the supported specific process. A PubMed search found no Neurospora-specific study of NCU08595; support rests on the fungal orthologs. Participation is as a non-catalytic assembly scaffold, not nuclease chemistry.

Neurospora cryo-EM structure 7R81 identifies this exact protein as eL39, entity 74/chain VB (author n1), in the translating large ribosomal subunit. The eL39-specific domain agrees with the direct structural assignment. GOA already provides cytosolic large ribosomal subunit (GO:0022625), a cellular-component descendant of ribosome. The emitted ribosome prediction is therefore correct but less precise than the established compartment and subunit assignment.

Neurospora cryo-EM structure 7R81 identifies this exact protein as eL39, entity 74/chain VB (author n1), in the translating large ribosomal subunit. The eL39-specific domain agrees with the direct structural assignment. GOA already provides cytosolic large ribosomal subunit (GO:0022625), a cellular-component descendant of ribonucleoprotein complex. The emitted ribonucleoprotein complex prediction is therefore correct but less precise than the established compartment and subunit assignment.

The SmG-specific InterPro signature (IPR034098) places this protein with characterized yeast SmG rather than the Lsm7 paralog, and yeast SmG is a subunit of the Sm ring that forms the core of the U1, U2, U4 and U5 snRNPs. Spliceosomal complex is correct, but the gene already carries exactly this term in GOA (InterPro2GO IEA, accepted in the gene review), together with more specific accepted complex terms such as U2-type prespliceosome (GO:0071004), so the prediction is not novel.

Primary metabolic process is correct but far less specific than the gene's existing annotation to mRNA splicing, via spliceosome (GO:0000398), an IBA from the eukaryotic METTL16 PAINT node (PTN000333540) that the gene review accepts; GO:0000398 is a descendant of GO:0044238 via RNA processing and RNA metabolic process. The underlying support is the conserved METTL16 U6 snRNA m6A activity shown experimentally for fission yeast Mtl16, which NCU11362 aligns to over its full length with the catalytic NPPF motif retained. PANTHER PTHR13393:SF0 alone would not justify this, because it also contains bacterial RlmF and PsiM; the transfer rests on the METTL16 PAINT node.

GLT-1 is the low-affinity D-glucose transporter of the N. crassa dual-affinity glucose uptake system; heterologous complementation of a hexose-transporter-null yeast and radiolabeled glucose uptake establish D-glucose transport (PMID:24581151, PMID:28115989). Transmembrane transporter activity is correct but less precise than the supported D-glucose transmembrane transporter activity and than the carbohydrate:proton symporter IBA the gene already carries in GOA, whose carbohydrate transport component is experimentally supported. The exact term is also already in GOA (InterPro IEA), so the prediction is not novel. This judgment concerns solute transport, not proton coupling.

GLT-1 is a twelve-helix MFS sugar porter that takes up extracellular glucose, and eGFP-tagged GLT-1 was imaged in both S. cerevisiae and N. crassa (PMID:28115989), supporting membrane residence; the supported location is the plasma membrane, inferred from the uptake function and the cell-periphery localisation of the same-subfamily relative MstE. Membrane is correct but less precise than that plasma-membrane assignment, and the exact term is already in GOA (IBA and IEA rows), so the prediction is not novel.

KAL-1 carries a canonical homeodomain (PROSITE PS50071, Pfam PF00046; DNA_BIND 77-136 with the WFQNRR recognition-helix motif) and sits in PANTHER subfamily PTHR24324:SF5, whose PAINT node PTN001963821 transfers RNA polymerase II cis-regulatory region sequence-specific DNA binding (GO:0000978) from experimentally characterised Phx1, Yox1, Pho2 and HHEX. Generic DNA binding is therefore correct, but it is already in GOA and is an ancestor of the curated GO:0000978 annotation that the gene review accepts, so it is less precise than the existing supported term. No KAL-1 DNA-binding experiment exists; PubMed has no functional study of KAL-1 or NCU03593.

Tagged VTC-4 co-localizes with the vacuolar ATPase subunit VMA-1 in prevacuolar compartments and the tubular vacuolar network of N. crassa hyphae (PMID:26453652 abstract; primary excerpts), which directly supports residence of this multi-pass VTC subunit on vacuolar membranes. The exact term is already in GOA for the gene (IEA), and the gene also carries its is_a descendant fungal-type vacuole membrane (GO:0000329, IBA), which the gene review accepts as the core location. The prediction is therefore correct but less specific than an existing supported annotation.

B8BAB0 has no GOA annotations, so the term would be novel, but it is not supported. B8BAB0 is 95.6% identical to japonica OsBURP12 (Q6ZA27) and sits in the PG1beta-like subfamily PTHR31458:SF2 with a signal peptide, the PG1beta FxxY/FTNYG repeat region and a C-terminal BURP domain; its closest PANTHER paralog is OsBURP16. The reproductive-development precedent in the BURP superfamily comes from RAFTIN, an anther-specific, cereal-restricted BURP clade (rice RAFTIN RNAi collapses pollen), and from the monocot BURP VII subfamily whose members are flower-predominant. The OpenScientist analysis places B8BAB0 outside that clade by full-length alignment (46.5% identity to tomato PG1beta versus 28.2% to wheat RAFTIN1A). The only gene-specific experiment, on the japonica allele, links OsBURP12 to polygalacturonase activity, pectin content and low-temperature seed germination, not pollen. The gene review accepts only extracellular region. The pollen label is therefore best explained as a transfer across nonisofunctional BURP paralog clades. No study has tested OsBURP12 for a pollen phenotype, so the refutation rests on clade placement rather than a negative experiment.

Q6YYC5 carries a RING domain and the RGLG4 subfamily assignment, so generic protein ubiquitination is supported, and the gene already carries GO:0016567 by IBA. The linkage-specific refinement has no direct support. Lys63-linked chain formation in vitro was shown for Arabidopsis RGLG2 (PMID:17586653), in the RGLG1/RGLG2 branch, whereas Q6YYC5 is sister to RGLG4 by PANTHER placement and by the OpenScientist identity and neighbor-joining analysis. The RGLG3/RGLG4 clade has ligase activity and drives GRXS17 degradation with the E2 UBC30 (PMID:27497447), but no linkage assay exists for it. Degradative outcomes do not separate the branches, because RGLG1/RGLG2 also mediate proteasomal degradation of AtERF53 (PMID:22095047). Lys63 chain formation requires a Ubc13-type E2 heterodimer, so the RING domain alone cannot settle linkage. The PANTHER family review scores this term as unresolved and the gene review leaves linkage as an open question. The prediction is neither supported nor refuted; a linkage assay with the rice cognate E2 would be needed.

Q6YYC5 has the RGLG architecture with a C-terminal RING-type domain (residues 356-389) and is placed in PANTHER subfamily PTHR45751:SF16 with Arabidopsis RGLG4. RGLG3 and RGLG4 have experimentally demonstrated ubiquitin ligase activity (PMID:22898498), and RGLG4 with UBC30 promotes ubiquitination and degradation of GRXS17 (PMID:27497447), which justifies transfer of E3 activity within the subfamily as an inference rather than an observation on the rice protein. GOA carries only the parent term GO:0004842 ubiquitin-protein transferase activity (IBA and IEA), which GO:0061630 is_a; the gene review refines both rows to GO:0061630 and makes it the core molecular function. The prediction is therefore correct and more specific than the existing annotation, without any claim about substrate or chain linkage.

The target is the full-length bonobo ortholog of human SLC26A11 (99.2% identical, no gaps), with the SLC26/SulP transmembrane domain and a C-terminal STAS domain. SLC26A11 conducts chloride: the human protein shows an uncoupled, channel-like chloride conductance (PMID:42509233), and mouse KBAT functions as a constitutively active chloride channel (PMID:23733100). Chloride conduction supports monoatomic anion transmembrane transporter activity; sulfate, the other substrate, is polyatomic and does not. The gene already carries the more specific supported descendant chloride channel activity (GO:0005254), so the prediction is correct but less precise.

The target has multiple predicted transmembrane segments, so membrane residence is supported by sequence architecture. Human SLC26A11, its one-to-one ortholog, localizes to lysosomes in mammalian cell lines (PMID:42509233), and the gene already carries the supported lysosomal membrane term (GO:0005765), an is_a descendant of membrane. Membrane is correct but less precise.

The protein kinase domain (96-348) carries the PROSITE protein-kinase ATP-binding and serine/threonine-kinase active-site signatures, and motif analysis finds the glycine-rich loop, beta3 lysine K125, HRD catalytic loop, DFG and APE motifs all intact, so the protein is predicted to be a catalytically competent kinase rather than a pseudokinase. The gene review accepts protein serine/threonine kinase activity (GO:0004674) as its core function on this basis. An active protein kinase performs the phosphotransfer step of protein phosphorylation itself, so the process term follows directly from the supported activity. GO:0006468 is not among the gene's GOA annotations, and OLS4 lists no is_a or part_of path from GO:0004674 to it, so the prediction adds a novel process annotation. Support is sequence-based inference; no substrate or phosphorylation event has been observed, and the term implies nothing about the autophagy roles that the gene review marks over-annotated.

The target is a full-length small heat-shock protein with an alpha-crystallin domain. The PTHR11527 family-root PAINT node PTN000163021 carries response to salt stress, and the closest placed relative of B7FXQ8 is a Thalassiosira pseudonana PTHR11527:SF175 sHSP. The family-root salt IBD rests on stimulus-specific plant phenotypes, and no salt induction, protection or loss-of-function data exist for B7FXQ8 or any characterised protein in its diatom lineage. The gene review treats this IBD as an open question rather than transferring it, and the OpenScientist analysis flags the salt prediction as an outlier relative to the family's InterPro2GO mapping. Characterised sHSP client binding (PMID:29983375) does not establish a salt-triggered response. The term is absent from GOA for this gene, and the prediction can neither be supported nor refuted.

B7FXQ8 is a full-length sHSP with a complete alpha-crystallin domain and retains the C-terminal I-X-I oligomer-assembly motif (IAI, residues 150-152). Protein complex oligomerization is a PAINT IBD at the PTHR11527 family root PTN000163021, from which the diatom SF175 sHSPs that group with B7FXQ8 descend, and oligomer assembly is demonstrated experimentally for sHSPs such as archaeal Hsp16.5 (PMID:29983375). This is a justified family-level inference, not an observation on the target, and it does not fix an oligomer size. The gene review proposes this process as a new annotation. B7FXQ8 has no GOA annotations, so the prediction is novel.

Protein folding is among the PAINT IBDs at the PTHR11527 family root, and characterised sHSPs bind non-native clients and suppress their aggregation (PMID:29983375), which can feed into ATP-dependent refolding by other chaperones. The re-reviewed gene and its bioinformatic analysis limit transfer to holdase client binding, oligomerization and response to heat, and treat sHSPs as holdases that do not themselves assist folding; the PTHR11527 family review leaves the protein-folding process unresolved. No folding, refolding or aggregation assay exists for B7FXQ8 or any protein in its diatom lineage. The term is absent from GOA for this gene. Whether holdase activity amounts to participation in protein folding for this protein is not settled by the available evidence.

Response to heat is a PAINT IBD at the PTHR11527 family root PTN000163021, and in an alpha- crystallin-domain tree B7FXQ8 groups with a Thalassiosira pseudonana PTHR11527:SF175 sHSP that descends from that node. B7FXQ8 is a full-length member with a complete alpha-crystallin domain, and characterised sHSPs bind unfolded clients (PMID:29983375), the mechanism by which they hold heat- denatured proteins. The gene review proposes response to heat as a new annotation on this basis. The support is a phylogenetic transfer; no heat-induction or thermotolerance data exist for B7FXQ8 itself. B7FXQ8 has no GOA annotations, so the prediction is novel.

The small-HSP architecture supports a holdase chaperone role, and response to hydrogen peroxide is a PAINT IBD at the PTHR11527 family root. That IBD rests on stimulus-specific phenotypes in other branches, and the gene review leaves its family-wide applicability as an open question rather than transferring it. No peroxide induction, protection or perturbation data exist for B7FXQ8 or a characterised protein in its diatom lineage, and the OpenScientist analysis flags the peroxide prediction as an outlier. The term is absent from GOA for this gene and can neither be supported nor refuted.

B7FXQ8 has the complete alpha-crystallin domain and I-X-I motif of small heat-shock proteins. Client-binding experiments on characterised sHSPs show recognition of partially or globally unfolded proteins (PMID:29983375), and the gene review's core function is an ATP-independent holdase that binds non-native clients. This supports the binding claim as a family-level inference. GO:0051082 is obsolete in the current ontology without a replacement; the gene review records the activity under a proposed holdase chaperone term. The biological claim is correct, but the term is unusable for a live annotation. B7FXQ8 has no GOA annotations, so the claim is novel.

The target has a UBC conjugating-enzyme domain and the active-site cysteine annotation, consistent with its UBE2L6-family assignment. Experiments on UbcH8/UBE2L6 establish E2-mediated ubiquitin and ISG15 conjugation (PMID:15131269), supporting transfer of the broad catalytic class. Cached target annotations already include ISG15 transferase and ubiquitin-like protein transferase activities. Generic transferase activity is supported but less precise than those functions.

Metal ion binding is supported by the target's zinc-finger architecture. The reviewed sequence record identifies a PHD finger at residues 122-170 (PROSITE-ProRule PRU00146) and an atypical RING finger at residues 18-79 (PRU00175). The PHD sequence retains the full Cys4-His-Cys3 zinc-ligand pattern (C124, C126, C141, C144, H149, C152, C165, C168), so zinc coordination by this domain is a diagnostic structural inference. No direct zinc-binding assay exists for the fission-yeast protein, and PubMed searches for the gene and the PHRF1 family returned no metal-binding experiment. The gene's GOA annotations contain no metal-ion or zinc-ion binding term (only ubiquitin protein ligase activity, which is not an is_a ancestor or descendant of metal ion binding), so the prediction is novel. The ProtNLM2 output was hydrated from GO:0008270 zinc ion binding, which would be the more precise term.

Purified fission-yeast Cbh1 directly binds centromeric K-type repeat DNA and was characterized by DNA-binding and footprinting assays. Its in vivo centromeric-chromatin association was independently examined. These target observations directly establish nucleic acid binding. The target already has the more precise experimental centromeric DNA-binding annotation GO:0019237, which is_a DNA binding and nucleic acid binding, and the exact predicted term is already carried by InterPro mapping. The broad prediction is therefore correct but not novel and less precise.

Budding-yeast Crt10 co-purifies with the cullin Rtt101 and binds it through the DDB1-like linker Mms1, and Rtt101-Mms1 is described as a CUL4(DDB1)-like ligase. GO, however, represents that assembly as the Cul8-RING ubiquitin ligase complex (GO:0035361), a sibling of the predicted Cul4-RING complex (GO:0080008) under cullin-RING ubiquitin ligase complex, and the budding-yeast review uses the Cul8 term. No study has tested whether fission-yeast Crt10, which is half the length of its budding-yeast counterpart, associates with any cullin ligase, including Pcu4-Ddb1 (Cul4); PubMed has no fission-yeast Crt10 papers. The gene review asserts no complex membership and notes that PomBase did not transfer the budding-yeast ligase-complex membership. The ProtNLM2 claim is a low-scoring phmmer transfer from Arabidopsis WDR5B Q9SY00 (model_score 0.11; phmmer_score 66.0), a WD-repeat protein outside the Crt10 family, so the source does not justify a family transfer. The complex claim is plausible by analogy but neither validated nor refuted.

No experimental annotation or publication links fission-yeast or budding-yeast Crt10 to UV-B responses; PubMed searches for Crt10 with UV or ultraviolet return nothing, and the gene review carries no DNA-damage or radiation-response term. The source is a TMalign structural comparison to Arabidopsis DHU1 Q8GYY7 (model_score 0.11; chain-normalized scores 0.56569 and 0.36277), so shared WD-repeat fold, not a conserved UV-B mechanism, underlies the claim. The characterized budding-yeast Crt10 directs nonfunctional 25S rRNA decay and is dispensable for the DNA-repair function of the Rtt101 ligase, which argues against but does not test a UV-B response. With no target evidence and no justified transfer, the claim is unresolved.

Fission yeast crystallography and biochemical reconstitution place Lsm6 in the Lsm5/6/7 subcomplex and in the Lsm1-7 ring bound to RNA, so membership of an Sm-like protein family complex is correct. The identical term GO:0120114 is already annotated to Q9UUI1 in GOA (IEA, ARBA), and the gene also carries the more specific Lsm2-8 and Lsm1-7-Pat1 complex terms, so the prediction is not novel.

DNA endonuclease activity is supported by fission yeast nuclease-mutant experiments, including the Rad32 nuclease requirement for endonucleolytic Rec12 removal, and by conserved Mre11 biochemistry. GOA already includes single-stranded DNA endonuclease activity (GO:0000014), in addition to this exact broad term. The prediction loses substrate specificity relative to that established annotation. This biological overlap does not establish training-set membership.

The conserved Mre11 nuclease domain supports 3′-to-5′ exonuclease chemistry, supported by purified-protein biochemistry and fission yeast catalytic-domain structure. GOA already records double-stranded DNA 3′-5′ DNA exonuclease activity (GO:0008311) by curated orthology transfer. The predicted term omits the substrate specificity of that established function. The net polarity of DNA-end resection does not refute the intrinsic exonuclease polarity.

Fission yeast rad32 mutants impair meiotic recombination, and Rad32 nuclease activity is required for removal of Rec12-linked ends. These experiments establish participation in the meiotic cell cycle. GOA already captures reciprocal meiotic recombination and double-strand break repair involved in meiotic recombination, which are more informative than the broad cell-cycle term. The prediction does not imply a general requirement for Rec12-dependent break formation.

Pta1 is a component of the experimentally characterized fission yeast CPF core. Purified-factor rescue of budding yeast pta1 mutant extracts establishes its conserved participation in mRNA cleavage and poly(A) addition. GOA already records co-transcriptional mRNA 3′-end processing, cleavage and polyadenylation pathway (GO:0180010). The broad mRNA-processing prediction is correct but loses the established mechanistic specificity and does not assign intrinsic nuclease activity.

The rfc3-1 temperature-sensitive mutant is defective in DNA replication (PMID:10588638), and purified fission-yeast RFC is required with PCNA for processive DNA synthesis by Pol delta (PMID:10748208). Replication is an established function of this clamp-loader subunit. The gene already carries DNA replication itself in GOA (IEA and NAS), so the prediction is correct and not novel (CNN). The gene review refines that row to the more specific DNA-templated DNA replication.

S. pombe RPA51/Rpa49 co-purifies with Pol I, including in stationary phase (PMID:12893961), and the A49/A34.5 subcomplex is present in purified S. pombe Pol I used for cryo-EM (PMID:33536435). Membership of a DNA-directed RNA polymerase complex is therefore supported. The gene already carries RNA polymerase I complex (GO:0005736, IDA and IBA), an is_a descendant of the predicted term, so the prediction is LSP. The protein is an accessory subunit, not a catalytic one.

S. pombe RPA51/Rpa49 is a Pol I subunit needed for maximal rDNA transcription in vivo and in vitro (PMID:12893961), so participation in DNA-templated transcription is supported. The exact term is already on the gene as an InterPro2GO IEA, which the gene review modifies to its specific descendants; the gene carries transcription elongation by RNA polymerase I (GO:0006362, IDA and IBA) and RNA polymerase I preinitiation complex assembly (GO:0001188, IBA). Because the prediction is materially less specific than these supported annotations, it is LSP rather than CNN.

Purified S. pombe Rpo41 binds mitochondrial promoters and supports transcription together with Mtf1 (PMID:21357609). The predicted DNA binding is biologically supported. The existing mitochondrial promoter sequence-specific DNA binding annotation is more specific within the same GO aspect, so the prediction is LSP.

Purified S. pombe Rpo41 binds mitochondrial promoters and supports transcription together with Mtf1 (PMID:21357609). The predicted DNA-templated transcription is biologically supported. The existing mitochondrial transcription annotation is more specific within the same GO aspect, so the prediction is LSP.

Purified S. pombe Rpo41 binds mitochondrial promoters and supports transcription together with Mtf1 (PMID:21357609). The predicted DNA-directed RNA polymerase complex is biologically supported. The existing mitochondrial DNA-directed RNA polymerase complex annotation is more specific within the same GO aspect, so the prediction is LSP.

The specific RRP36 domain and target nucleolar localization support the conserved preribosome role inferred by PAINT. Purification and depletion in budding yeast establish 90S/pre-40S association and early small-subunit rRNA maturation; human depletion supports conservation across eukaryotes. GOA already records maturation of SSU-rRNA from tricistronic rRNA transcript (GO:0000462). The broad rRNA-processing prediction is supported but loses small-subunit specificity; no intrinsic RNase chemistry is implied.

Rrp36 belongs to a specifically identified ribosome-biogenesis family; characterized ortholog copurification with RNA-containing 90S and pre-40S particles establishes the relevant complex type. Curated ancestral inference already places fission yeast Rrp36 in the 90S preribosome (GO:0030686), consistent with its nucleolar localization. A ribonucleoprotein complex is a broader description of that supported complex membership. The prediction is correct but less precise than the existing 90S assignment.

Sen15 belongs to the fungal Sen15 family (IPR042777) of noncatalytic endonuclease subunits; the cleavage active sites lie in Sen2 and Sen34, and Sen15 pairs with Sen34 as a structural partner. Participation in tRNA splicing is therefore correct. However, Sen15 contributes only to the endonucleolytic step, not to exon ligation or 2'-phosphate removal, and the gene already carries the more specific GO:0000379 tRNA-type intron splice site recognition and cleavage (IBA from the same PANTHER subfamily and InterPro2GO), a descendant of the predicted GO:0006388 (OLS4 hierarchical ancestors). The prediction is broadly correct but less specific than the supported existing annotation. No fission-yeast experimental study of sen15 was found in PubMed.

Mutant, localization and biochemical experiments in characterized budding yeast Sus1 establish an mRNA-export role through Sac3-Thp1/TREX-2. The fission yeast protein belongs to the conserved Sus1/ENY2 family, and curated phylogenetic and orthology annotations support the same complex role. GOA already records the more specific poly(A)+ mRNA export from nucleus (GO:0016973). The prediction is biologically supported through that justified comparative inference, with less transcript specificity.

Fission yeast purification places Sus1 in SAGA, and characterized budding yeast Sus1 mutants establish participation in histone H2B deubiquitination and maintenance of histone H3 methylation. Transfer is justified by conserved Sus1 family identity and SAGA/DUB-module context; this is a noncatalytic contribution to chromatin regulation. GOA already specifies transcription initiation-coupled chromatin remodeling (GO:0045815). Generic chromatin organization is therefore supported but less precise.

The target has the diagnostic uS13m family domain and mitochondrial localization. Characterized S. cerevisiae mitoribosome structures support transfer of the conserved mitochondrial ribosomal role. Translation is therefore supported, and the existing mitochondrial translation annotation (GO:0032543) is more specific within the biological-process aspect. The generic prediction is LSP.

The target has the diagnostic uS13m family domain and mitochondrial localization. Characterized S. cerevisiae mitoribosome structures support transfer of the conserved mitochondrial ribosomal role. Membership in the mitochondrial small ribosomal subunit is a supported cellular-component annotation and is more specific than ribosome. The predicted broad complex membership is LSP.

The target carries the diagnostic uS13 domain. In the experimentally determined S. cerevisiae mitoribosome, the homolog Sws2/P53937 is uS13m and contacts rRNA helix H62 (PMID:28154081, supplementary tables S2 and bridge table). This supports transfer of conserved RNA binding to fission-yeast Sws2, without asserting identical bridge geometry. The equivalent RNA-binding annotation already exists in target GOA, so CNN applies.

The target has the diagnostic uS13m family domain and mitochondrial localization. Characterized S. cerevisiae mitoribosome structures support transfer of the conserved mitochondrial ribosomal role. Membership in the mitochondrial small ribosomal subunit is a supported cellular-component annotation and is more specific than ribonucleoprotein complex. The predicted broad complex membership is LSP.

Fission-yeast Uch2 is directly characterized as the major deubiquitinating enzyme associated with the 26S proteasome, and proteomic work shows it co-purifies the complete proteasome. Processing ubiquitin conjugates at this complex is part of proteasome-mediated ubiquitin-dependent protein turnover, even though the DUB removes ubiquitin rather than degrading the substrate polypeptide itself. The predicted term is not novel; GOA already carries it (InterPro2GO), and the gene review treats it as correct but general, refining it to GO:0043161 proteasome-mediated ubiquitin-dependent protein catabolic process. GOA also carries the descendant GO:0071629 cytoplasm protein quality control by the ubiquitin-proteasome system (IGI, kept as non-core), an is_a descendant of GO:0006511 per OLS4. The prediction is therefore correct but less precise than supported annotations; viable deletion phenotypes reflect redundancy among proteasomal DUBs and do not negate participation.

Purified fission-yeast Ulp2 directly deconjugates SUMO from high-molecular-weight species, and the study identifies it as a cysteine protease. SUMO is a ubiquitin-like modifier, so the broad ubiquitin-like protein peptidase term is biologically correct. The more specific deSUMOylase activity GO:0016929 is experimentally established and already annotated; the XML records hydration from that term. This prediction loses substrate specificity and is therefore LSP. It does not imply a demonstrated ability to remove ubiquitin itself.

Direct biochemical experiments establish Ulp2 as a SUMO-deconjugating cysteine protease, including sensitivity to N-ethylmaleimide. The conserved C48 catalytic domain is consistent with that mechanism. Cysteine-type peptidase activity is true but less precise than the already annotated deSUMOylase activity and its SUMO substrate specificity.

Fission yeast Aps1/Vas2 deletion and pull-down experiments establish AP-1 assembly and exit transport from endosomes, including trafficking of the membrane protein Syb1. GOA already records endosome-to-Golgi and endosome-to-plasma-membrane protein transport. The predicted intracellular protein transport is correct but less precise than those established routes. The unrelated ARBA intercellular-transport assertion means movement between cells and is not evidence supporting this assessment. GO:0099638 endosome to plasma membrane protein transport is_a descendant of GO:0006886 (OLS4 ancestry), and the gene review accepts that specific IDA term as the refinement of the broad InterPro-mapped protein transport row.

The prediction is absent from GOA, which carries only nucleus (GO:0005634) for the cellular-component aspect, and the gene review accepts nucleus as the sole supported location. The reviewed UniProt record places Wss1 in the nucleus from experimental evidence, and the ProtNLM2 source hydrates the membrane claim from KW-0539 (Nucleus), which is not a membrane observation. The closest relative evidence, from budding-yeast Wss1 in the same PANTHER subfamily, is a GFP focus near the nuclear membrane and vacuolar localization upon genotoxic stress; proximity to the nuclear envelope and relocation to the vacuole do not demonstrate membrane association. Because a soluble protease lacking transmembrane helices could still associate peripherally with membranes, the claim is not refuted. Membrane fractionation, nuclear-envelope colocalization or a defined membrane-associated partner complex for Wss1 is missing.

The target has the diagnostic uL4m family domain and mitochondrial localization. Target Yml6 also co-purifies at low levels with Ppr10 among mitoribosomal proteins (PMID:34119521). Characterized S. cerevisiae mitoribosome structures support transfer of the conserved mitochondrial ribosomal role. Translation is therefore supported, and the existing mitochondrial translation annotation (GO:0032543) is more specific within the biological-process aspect. The generic prediction is LSP.

The target has the diagnostic uL4m family domain and mitochondrial localization. Target Yml6 also co-purifies at low levels with Ppr10 among mitoribosomal proteins (PMID:34119521). Characterized S. cerevisiae mitoribosome structures support transfer of the conserved mitochondrial ribosomal role. Membership in the mitochondrial large ribosomal subunit is a supported cellular-component annotation and is more specific than ribonucleoprotein complex. The predicted broad complex membership is LSP.

The target has the diagnostic uL4m family domain and mitochondrial localization. Target Yml6 also co-purifies at low levels with Ppr10 among mitoribosomal proteins (PMID:34119521). Characterized S. cerevisiae mitoribosome structures support transfer of the conserved mitochondrial ribosomal role. Membership in the mitochondrial large ribosomal subunit is a supported cellular-component annotation and is more specific than ribosome. The predicted broad complex membership is LSP.

C6T1A2 is a single-finger Q-type C2H2 zinc-finger protein in PTHR47593:SF9. Arabidopsis ZFP3 and related ZFPs confer ABA insensitivity at germination when overexpressed (PMID:24808098), and the PAINT GO:0009788 node covering Arabidopsis ZFP4/ZFP7 lies in the PTHR45730 tree, which does not contain C6T1A2; those proteins are only about 34-37% identical to it. The soybean QALGGH protein GmZFP3 (a chromosome 7 locus, not C6T1A2) acts negatively in drought tolerance and may act through an ABA-dependent pathway (PMID:27047508), which is not evidence for this gene. No experiment on C6T1A2 or any SF9 member tests ABA signaling, and no rows exist in GOA, so the prediction is plausible as a family-class hypothesis but cannot be validated or refuted.

C6T1A2 has one Q-type C2H2 zinc finger (QALGGH, residues 92-97) and was cloned in a soybean transcription-factor ORFeome, so a nuclear location is expected from domain architecture. It has not been tested for C6T1A2 or any PTHR47593:SF9 member. Arabidopsis ZFP3 is nuclear (PMID:24808098) but lies in a different PANTHER tree, so this is class-level context rather than a phylogenetic transfer, and the PAINT nucleus node (PTN009184003) is likewise outside the C6T1A2 tree. A soybean QALGGH relative, GmZFP3, was distributed throughout the cell rather than confined to the nucleus (PMID:27047508), so domain class alone does not fix the localization. Nucleus is absent from GOA for this gene. The prediction is plausible but supported only by domain inference.

Q9KZ33 carries the ECF sigma-70 domain signature (IPR052704) and the sigma3/sigma4 superfamily, and its gene review accepts the PAINT IBA for sigma factor activity (node PTN001249270 of PTHR30173, with the target in subfamily SF43 alongside M. tuberculosis ECF41 SigI). Sigma factors do part of the work of initiation itself: they recognise the promoter and melt it as the initiating subunit of RNA polymerase holoenzyme, so participation in DNA-templated transcription initiation follows from the supported activity by family-wide transfer; the PTHR30173 family review grants this term family-wide. The gene's GOA set has sigma factor activity and regulation of DNA-templated transcription initiation (GO:2000142); OLS4 shows GO:2000142 is not an is_a/part_of descendant of GO:0006352, so the prediction is neither present nor less specific than an existing annotation. No experiment on SCO7099 addresses its transcriptional role (a PubMed search for SCO7099 returns no dedicated study), so this rests on the phylogenetic and domain inference rather than direct observation.

GOA has no annotations for SCO2678, so the term is novel. The sole family signature is Pfam PF18143 (HAD_SAK_2), and a reproducible motif scan finds all four HAD catalytic motifs, which supports a generic Mg2+-dependent phosphatase activity (the gene review's only core function). PF18143 was defined from eukaryotic RNA-repair-associated proteins, no member has been characterized, and the family is not one of the HAD subfamilies containing experimentally characterized 5'-nucleotidases, so there is no justified phylogenetic transfer of nucleotide substrate specificity. No PubMed record describes SCO2678 or a characterized ortholog. The prediction is a plausible refinement of the supported phosphatase activity, but no evidence establishes or excludes nucleotides as substrates, so it remains uncertain.

GOA has no annotations for SCO2678, so the term is novel. Participation in deoxyribonucleotide catabolism would require dNMP hydrolysis, which is itself unsupported: the protein is a stand-alone HAD_SAK_2 (PF18143) domain with no characterized family member and shares no Pfam family with HAD enzymes shown to dephosphorylate dNMPs. Its KEGG neighbourhood contains no nucleotide-catabolism genes (and no RNA-repair genes either), and PubMed has no study of SCO2678. The OpenScientist analysis reaches the same evidential picture. Absence of supporting evidence does not refute the process, because the substrate of this HAD phosphatase is unknown, so the prediction is uncertain rather than incorrect.

Correct but not novel. The target carries the GAS7-specific F-BAR domain (residues 121-381) and sits in PANTHER subfamily PTHR23065:SF57, inside the PAINT node PTN002747256 whose IBD places cytoplasm across F-BAR proteins; F-BAR proteins are cytosolic and associate peripherally with membranes, as shown for GAS7 assembling on membranes during phagocytic-cup formation (PMID:31628328). The exact term GO:0005737 cytoplasm is already in GOA for this gene (TreeGrafter IEA), and the gene review keeps it as a correct, non-core localization; the functional site is the cytoplasmic face of the plasma membrane, which is also already annotated.

The target is a catalytically intact PRORP in PANTHER subfamily PTHR13547:SF13 with Arabidopsis PRORP2, and the gene already carries tRNA 5'-leader removal (GO:0001682, IBA and IEA), which the gene review accepts. Arabidopsis PRORP2 and PRORP3 mature nuclear pre-tRNA 5' ends in vivo (PMID:26655022) and PRORPs supply RNase P activity in plants (PMID:22549728), justifying the subfamily transfer. GO:0008033 tRNA processing is an is_a ancestor of GO:0001682 (OLS4), so the prediction is correct but less specific than the existing supported annotation. No tobacco-specific experimental literature was found in PubMed.

The 129-residue target contains a KilA-N domain spanning residues 19–124 and a KilA/APSES helix-turn-helix assignment. The primary comparative analysis defining KilA-N identifies a conserved DNA-binding module in viral and cellular regulatory proteins (PMID:11897024). This provides a domain-based inference for broad DNA binding, not an experimental binding measurement on Trichomonas A2FPI7 or evidence for a specific DNA sequence. The molecular function is absent from the cached annotations.

Patatin-family acyl-ester hydrolysis depends on a Ser-Asp catalytic dyad, and mutation of the serine abolishes esterase activity (PMID:12779324). The deposited 302-residue model lacks the N-terminal catalytic region, and the gene-model analysis shows that this is not only a truncation artifact: the upstream genomic sequence that probably encodes the true N-terminus and the full-length 3B homoeolog both lack the G-T-S-T-G nucleophile elbow and the canonical DGGG oxyanion block, while intact paralogs in the same chromosome-3 cluster retain them. The protein is therefore predicted to be a non-catalytic patatin-like protein, and lipase activity is refuted. The gene review removes the electronic hydrolase activity annotation on these grounds. The lipase IBAs in the UniProt record propagate from a family-level PAINT node and do not detect this target-specific loss of the catalytic serine. No experimental data exist for this protein.

Lipid catabolic process is not carried by this gene in GOA; the gene has only the broader InterPro2GO lipid metabolic process annotation, which the gene review marks as over-annotated. Patatin-family membership would place the protein in lipid catabolism through acyl-ester hydrolysis, but the locus lacks the catalytic serine nucleophile and canonical oxyanion glycines, so hydrolysis by this protein cannot be assumed. A non-catalytic participant (for example a lipid-binding or regulatory role alongside the active paralogs in the same cluster) could still be involved in the process, so the absence of catalysis does not refute the term. No expression, interaction or phenotype data exist for this protein, and no literature on it was found, so the prediction can be neither validated nor refuted.

The target is in PANTHER subfamily PTHR38710:SF1 together with Arabidopsis GLCAK1, which was shown with recombinant protein to phosphorylate D-glucuronic acid to D-glucuronate-1-phosphate using ATP (PMID:19951951). The wheat protein keeps both GHMP kinase domains and the glucuronokinase-like signature, so the transfer of ATP-dependent sugar phosphorylation is a justified phylogenetic inference. The gene review proposes glucuronokinase activity (GO:0047940) and UDP-glucuronate biosynthetic process as NEW annotations on the same basis. A wheat study overexpressed the three hexaploid-wheat GlcAK homoeologs in Arabidopsis and reported metabolic phenotypes, but its abstract reports no kinase assay (PMID:36931562). GOA carries no phosphorylation term or descendant for this gene; it has only cytosol, ATP binding and a mevalonate-pathway mapping that the gene review removes. The prediction is therefore novel. It is broad, and a more informative annotation would name the glucuronate substrate.

The target is in PANTHER subfamily PTHR12480:SF21 with Arabidopsis JMJ22 (Q67XX3), the reviewed protein named as the phmmer hit in the ProtNLM2 evidence record; the two share an F-box and JmjC domain, 62.6% global identity and the conserved Fe(II)-binding His330/Asp332/His413 triad. In Arabidopsis, the photomorphogenesis-related annotation of JMJ22 comes from the phytochrome B-controlled germination study, in which JMJ22 acts redundantly with the SF6 paralog JMJ20 and is regulated by the SOMNUS repressor (PMID:22483719); the JMJ22 review keeps this term as non-core. The wheat gene review asserts no light-regulated process, and a PubMed search found no functional study of the wheat protein or any cereal JMJ22 ortholog. Orthology supports a conserved nuclear JmjC oxygenase but does not establish that the wheat protein sits in an equivalent light-signalling circuit, so the term can be neither confirmed nor refuted. It is absent from the target's GOA annotations.

The target is in PANTHER subfamily PTHR12480:SF21 with Arabidopsis JMJ22 (Q67XX3), the reviewed protein named as the phmmer hit in the ProtNLM2 evidence record; the two share an F-box and JmjC domain, 62.6% global identity and the conserved Fe(II)-binding His330/Asp332/His413 triad. The only experimental basis for this term in the family is the JMJ22 IMP annotation, and the JMJ22 evidence places the protein upstream of gibberellin biosynthesis: JMJ22 removes H4R3me2 at GA3ox1/GA3ox2 and so raises gibberellin levels (PMID:22483719). That is regulation of hormone synthesis, not participation in gibberellin perception or signal transduction, and the reviewed JMJ22 record describes the role as triggering GA biosynthesis. The Arabidopsis JMJ22 gene review accordingly replaces this term with positive regulation of gibberellin biosynthetic process. The prediction therefore transfers a mis-specified process from the ortholog; even if the wheat protein shares the JMJ22 role, gibberellin-mediated signalling is the wrong process. No wheat experiment addresses it, and the term is absent from the target's GOA annotations.

The target is in PANTHER subfamily PTHR12480:SF21 with Arabidopsis JMJ22 (Q67XX3), the reviewed protein named as the phmmer hit in the ProtNLM2 evidence record; the two share an F-box and JmjC domain, 62.6% global identity and the conserved Fe(II)-binding His330/Asp332/His413 triad. JMJ22 is an experimentally characterised H4R3me2 arginine demethylase that activates GA3ox1/GA3ox2 expression by removing a repressive histone mark (PMID:22483719), and the JMJ22 gene review accepts epigenetic regulation of gene expression as its core process. Because the catalytic site and domain architecture are conserved in the closest same-subfamily ortholog, transfer of this broad chromatin-level process is a justified family inference, though not an observation on the wheat protein. The transfer has limits: the histone substrate has not been tested in wheat, the wheat gene review holds back from asserting a histone demethylase activity, and Arabidopsis JMJ22 also affects SAC51-family mRNA stability (PMID:35962471), so not every JMJ22 function is chromatin-based. The term is absent from the target's GOA annotations, which carry only nucleus, DNA binding and metal ion binding.

The target is in PANTHER subfamily PTHR12480:SF21 with Arabidopsis JMJ22 (Q67XX3), the reviewed protein named as the phmmer hit in the ProtNLM2 evidence record; the two share an F-box and JmjC domain, 62.6% global identity and the conserved Fe(II)-binding His330/Asp332/His413 triad. Arabidopsis JMJ22 is derepressed when phytochrome B is activated by red light (PMID:22483719), and the JMJ22 gene review keeps response to red light as a non-core term describing how JMJ22 is regulated. That regulation depends on the SOMNUS-phyB circuit, which is not shown for the wheat gene; conservation of the catalytic domain does not establish conservation of its transcriptional control, and no wheat expression or genetic study was found by PubMed search. The prediction cannot be confirmed or refuted and is absent from the target's GOA annotations.

The target is in PANTHER subfamily PTHR12480:SF21 with Arabidopsis JMJ22 (Q67XX3), the reviewed protein named as the phmmer hit in the ProtNLM2 evidence record; the two share an F-box and JmjC domain, 62.6% global identity and the conserved Fe(II)-binding His330/Asp332/His413 triad. Arabidopsis JMJ22 promotes light-induced seed germination redundantly with the SF6 paralog JMJ20, by raising gibberellin synthesis at GA3ox1/GA3ox2 (PMID:22483719), and the JMJ22 gene review accepts this as its best-established biological role. Whether the wheat ortholog has the same role depends on its expression in seed, its target loci and the regulatory context of cereal germination, none of which has been tested; PubMed searches found no functional study of the wheat gene. A related germination role is plausible but not established, and the term is absent from the target's GOA annotations.

F6LAX4 carries the PP2A regulatory subunit A family signatures (IPR051023, PANTHER PTHR10648:SF4) and descends from the eukaryote PP2A scaffold PAINT nodes, so the conserved scaffold-catalytic subunit association transfers to it as a family inference. Structural work on the PP2A holoenzyme shows that the scaffolding A subunit and the catalytic C subunit form a heterodimeric core enzyme to which a regulatory B subunit then binds (PMID:17174897), which is binding to a nonidentical protein to form a heterodimer. The term is not in the gene's GOA set and is not an ancestor or descendant of the existing protein phosphatase regulator activity IBA or of the protein complex scaffold activity the gene review proposes, so it is novel. It is a less informative description of the A subunit than complex scaffolding, but it is correct. The OpenScientist analysis judged the term NPI because the scaffold nucleates an A-B-C heterotrimer, but the A-C core enzyme it forms is itself a heterodimer, so the term is correct though less informative than a scaffold term.

The target is a Triticum aestivum protein. Neuronal cell body denotes the soma of a neuron, a cell type absent from plants, so this compartment cannot apply to the wheat PP2A scaffold. The term is absent from the gene's annotations and nothing in the gene review or the PTHR10648 family review supports it for a plant member. The prediction alone does not establish which training example or model mechanism produced the error.

In Arabidopsis, the PP2A regulatory subunits B'alpha and B'beta are required to maintain centromeric sister chromatid cohesion in meiosis, and the A1 structural subunit, a co-ortholog of F6LAX4, localizes with them to meiotic centromeres (PMID:30061120). This places the wheat scaffold plausibly in a PP2A-B' holoenzyme acting on chromosomes. The demonstrated function, however, is protection of centromeric cohesion, which GO represents as a sister chromatid cohesion process outside chromosome segregation; segregation defects in the mutants are a downstream consequence, and the genetic evidence is on the B' subunits, not on an A subunit. PAINT placed its meiotic centromeric cohesion IBD on an animal/fungal subclade that does not include F6LAX4. Support for the exact segregation term is therefore insufficient. The OpenScientist analysis judged the term NPI because the Shugoshin-PP2A cohesion role is mediated by the B56 regulatory subunit; the call here is UNC rather than NPI because the A1 co-ortholog sits in that B' holoenzyme at meiotic centromeres, so a holoenzyme contribution is plausible but not shown for segregation itself.

GFP-tagged Arabidopsis PP2AA1, the structural A subunit and a co-ortholog of F6LAX4 (grasses have a single A-subunit lineage co-orthologous to all three Arabidopsis A subunits), colocalizes with centromeric histone CENH3 in male meiocytes from pachytene to metaphase II, together with the B'alpha regulatory subunit (PMID:30061120). Centromere targeting depends on the regulatory subunit rather than on the scaffold itself, but the A subunit is a member of the holoenzyme residing there, which suffices for a cellular component annotation. Transfer to the wheat ortholog is an inference, restricted in the source to meiotic stages. The term is absent from the gene's GOA set and is novel. The OpenScientist analysis judged the term NPI because centromere targeting is B56-mediated, but the colocalization of the A1 co-ortholog with CENH3 shows the scaffold is present there as a holoenzyme member, which a cellular component term requires.

The target organism is Triticum aestivum. Neuron projections are neuronal structures, and plants have no neurons, so this localization is incompatible with the target taxon. It is absent from the gene's annotations and is not supported by the gene review or the PTHR10648 family review for a plant member. No inference about model frequency bias is needed to establish the incompatibility.

The wheat product is a PP2A A-subunit scaffold. The human PP2A A-small t structure establishes that the viral small t antigen binds the scaffold as a PP2A inhibitor (PMID:17608567); the viral protein's name does not make that interaction immunological antigen recognition, which is what protein antigen binding describes. Plants lack the antibody and antigen-receptor systems that carry this activity, and the term is absent from the wheat annotations.

Q8P365 belongs to the glutathione peroxidase family and the BtuE subfamily (PANTHER PTHR11592:SF40), and descends from the gammaproteobacterial PAINT node PTN000173904 that contains experimentally characterized E. coli BtuE (P06610). It conserves the peroxidatic Cys37 active site. Purified E. coli BtuE decomposes peroxides and btuE is induced by oxidative-stress elicitors (PMID:20621065), which supports transfer of a protective role in the oxidative-stress response. The gene already carries an accepted IBA for GO:0034599 cellular response to oxidative stress, an is_a descendant of the predicted GO:0006979 (OLS4 ancestry), and the gene review marks the GOA IEA row for GO:0006979 itself as TOO_GENERAL. The prediction is correct but less precise than the existing supported annotation. No experiment on the Xanthomonas protein itself was found in a PubMed search.

The 262-residue target has an N-terminal signal peptide (1-28) followed by a solute-binding protein family 3 domain covering residues 40-259, and the record has no transmembrane segments. This is the architecture of an exported periplasmic binding protein, not a pore-forming channel. Family-3 binding modules resemble the ligand-binding lobes of ionotropic glutamate receptors, but in those channels the module is fused to a transmembrane pore domain that D3VIU4 lacks. In the characterised E. coli cystine importer (FliY, a sister subfamily) the binding protein hands its ligand to separate membrane-spanning and ATPase partners (YecS, YecC), which is the role the gene review assigns to D3VIU4 alongside its neighbouring ATPase and permease genes. The gene review carries no channel annotation, the PANTHER PTHR35936 family review records that the target lacks the pore architecture required for this activity, and the OpenScientist investigation of this prediction reached the same conclusion from domain and structural analysis. Binding a transported ligand does not make the binding component a ligand-gated ion channel.

The target contains eight tandem C2H2 zinc-finger domains, consistent with sequence-specific DNA recognition. The cached GOA record has an exact curated IBA for RNA polymerase II cis-regulatory region sequence-specific DNA binding, alongside the corresponding transcription-factor activity. This is a curator-assessed phylogenetic inference consistent with the target architecture, not an ARBA assertion. The available evidence supports the function without identifying a particular DNA motif or target gene.

The target has the tandem C2H2 zinc-finger architecture of a sequence-specific DNA-binding factor. The exact regulation of transcription by RNA polymerase II annotation is present as a curated IBA in the cached GOA record, together with compatible DNA-binding transcription-factor activity. These phylogenetic annotations and the target architecture support the broad regulatory process. They do not specify whether the factor activates or represses its targets.

The target zinc-finger architecture is compatible with its curated DNA-binding transcription-factor assignment. Nucleus is already present as a curated IBA in the cached GOA record, supporting this localization through a phylogenetic judgment consistent with the molecular function. No target localization experiment is asserted. The exact overlap makes this a supported, non-novel annotation.

The zinc-finger array and existing curated IBA support a DNA-binding RNA polymerase II transcription factor. They do not determine whether its effect on transcription is activating, repressing, or context-dependent. Absence of a recognizable activation domain is not decisive because activation can depend on partners or short interaction regions. The activator-specific term is absent from the cached annotations, and evidence for regulatory direction is needed. The OpenScientist analysis judged the term NPI because the effector-less zinc-finger array cannot specify regulatory direction; that same argument leaves activation undetermined rather than excluded, so the call is UNC.

UNC. Neither J3KS48 nor the canonical NARF product Q9UHQ1 carries any iron-sulfur annotation in GOA, and no experiment has shown a bound Fe-S cluster in NARF, so the term would be novel but is unsupported. Hydrogenase-family (PTHR11615) homology makes cysteine-ligated cluster binding a hypothesis only; the family review keeps NARF/IOP2 distinct from NARFL/IOP1 and transfers no Fe-S term to NARF. A report of in vitro H2 production with a sequence-inferred [4Fe4S] cluster concerns recombinant IOP1/NARFL, not NARF, so it does not transfer across this paralog split. At the product level the required feature is unresolved. J3KS48 is a 217-residue product lacking much of the 456-residue reference, including Cys172, the C-terminal Cys374/Cys378 pair of the large-subunit-like domain and the whole small-subunit-like domain, so a complete hydrogenase-like cluster site is unlikely, but it keeps the N-terminal cysteine-rich region (12 of 15 cysteines), and no mapping establishes whether any cluster site in NARF depends only on the lost cysteines. The IOP2 knockdown result distinguishes CIA function but does not test cluster occupancy, so it neither validates nor refutes the claim.

F8WDQ9 lacks the reference UBE2F catalytic Cys116 and a complete UBC core. Moreover, the characterized full-length enzyme conjugates NEDD8, not ubiquitin. A UBC-domain match does not establish ubiquitin-protein transferase activity. Neither F8WDQ9 nor the canonical Q969M7 product carries this term in GOA, where the gene's experimental MF is NEDD8 conjugating enzyme activity; a PubMed search found no report of ubiquitin transfer by UBE2F.

F8WDQ9 is a 101-residue product that matches canonical Q969M7 residues 1-94 only. It lacks canonical residues 95-178, which hold the HPN motif and the catalytic thioester Cys116, so it covers only 70 of the 154 residues of the UBC core. The gene's known link to protein ubiquitination is indirect. Full-length UBE2F conjugates NEDD8 to cullin-5 through Cys116, and this activates CRL5 ubiquitin ligases. That catalytic route is not available to F8WDQ9. The product does keep the N-terminal UBA3-docking extension (residues 1-29) and the acetylated N-terminus that binds DCNL co-E3s, and UBE2F N-terminal peptides alone can inhibit DCNLs. A regulatory or competitive binding role in CRL5 activation, and so in ubiquitination, is therefore plausible but neither shown nor excluded. There is no positive refutation, so the call is uncertain. The term is absent from GOA for both F8WDQ9 and canonical Q969M7, whose process annotations are protein neddylation.

Protein-FAD linkage is the conserved role of the Sdh5/SDHAF2 family, shown experimentally for yeast and human orthologs, and the gene already carries this term on the canonical product Q8C6I2 (ISO/ISS, accepted in that review), so it would not be novel at the gene level. For A0A494B8X4 itself the claim is unresolved: the product keeps the transit peptide, the N-terminal helical bundle and the Gly78-equivalent residue, but lacks the C-terminal helices of the Sdh5 domain, and it has never been tested for folding, SDHA binding or flavinylation. The gene review for this product deliberately withholds the activity for the same reason. Without product-specific evidence the prediction can be neither confirmed nor refuted.

A0A494B8X4 already carries mitochondrial matrix in GOA (GO_REF:0000044), and the gene review accepts it; the canonical product carries the same term. The location is supported independently of that electronic row: the product retains the complete 27-residue mouse transit peptide identical to Q8C6I2, and Sdh5/SDHAF2 orthologs are conserved mitochondrial proteins acting on the matrix-facing SDHA subunit. The prediction is correct but not novel.

Electron transfer from succinate to ubiquinone is performed by assembled complex II (SDHA-SDHD). Sdhaf2 is an assembly factor that binds SDHA and promotes its covalent flavinylation before complex assembly; it does not transfer electrons and is not part of the mature complex. Loss of Sdh5 abolishes SDH-dependent respiration, which shows necessity, not participation. The canonical-product review marks every GOA row for this term (IBA, UniRule IEA, ISO) as an over-annotation on these grounds, so the prediction most likely reproduces that over-propagated annotation. The appropriate process terms for the gene are complex II assembly and protein-FAD linkage, and for this truncated, unassayed product even those are unconfirmed.

SPCS2 is a noncatalytic accessory subunit of the signal peptidase complex; SEC11A or SEC11C carries the catalytic Ser-His-Asp triad. Neither canonical mouse Spcs2 nor human SPCS2 carries a peptidase activity annotation, and the PANTHER family review limits the family role to complex participation. The selected product additionally lacks the transmembrane architecture of the subunit. Catalysis by the complex does not establish peptidase activity of this protein.

GO:0006465 is obsolete in current GO (replaced by GO:0051604 protein maturation); the process is carried at the gene level by canonical Spcs2 (Q9CYN2) as GO:0016485 protein processing, so the prediction is not novel for the gene. SPCS2 is a noncatalytic subunit; catalysis is performed by SEC11A or SEC11C, so no catalytic step is lost. Its structurally resolved contribution is through its transmembrane domains, which with those of the other subunits form the bilayer window. A0A140LHW5 shares only canonical residues 1-66, within the N-terminal cytoplasmic domain (canonical 2-86), and lacks both transmembrane helices (87-107 and 112-132) and the C-terminal cytoplasmic domain, so it cannot contribute that membrane structure. The known mechanism does not, however, account for the full SPCS2 contribution to the process, and a binding or scaffolding role of the retained cytosolic segment toward complex or translocon components is neither shown nor excluded. There is no positive refutation, and whether the 74-residue product exists as a stable protein is also unestablished, so the prediction cannot be validated or refuted for this product.

Canonical Spcs2 (Q9CYN2) carries GO:0005787 (IBA, IEA, ISO), accepted in its review on human structural and yeast genetic evidence, so the term is not novel for the gene. SPCS2 is an integral two-pass ER membrane subunit of the complex, and its structurally resolved role in the complex is through its transmembrane region, which with the other subunits forms the membrane window. A0A140LHW5 retains only the N-terminal cytosolic residues 1-66 of the 226-residue subunit, followed by an unrelated eight-residue tail, and lacks both transmembrane helices (canonical 87-107 and 112-132), so it lacks the membrane region through which SPCS2 is a component of the signal peptidase complex.

MTMR12 is cytoplasmic, as shown for tagged human 3-PAP in COS-7 cells, where it also draws MTM1 into the cytoplasm, and in muscle it sits with MTM1 at triads overlapping the sarcoplasmic reticulum marker RyR1. The rat gene already carries GO:0016529 sarcoplasmic reticulum, accepted in the gene review, and cytoplasm is an ancestor of that term (OLS4), so the prediction is correct but less precise than an existing supported annotation. The selected product differs from the canonical rat product only in its first 40 residues, so the gene-level localization applies to it; the computational signal peptide on that alternative N-terminus does not establish secretion.

The process claim depends on C28G1.2 acting as a serpin-type protease inhibitor. Local alignment to its inhibitory paralogs shows no counterpart to the hinge and reactive-centre loop, the element serpins use to trap target proteases, so the inhibitory function of the clade is not expected to hold for this divergent lineage. C. elegans serpins with demonstrated inhibitory activity are restricted to five full-length members, which do not include C28G1.2. No other mechanism of peptidase regulation has been reported for the gene; the conclusion is a sequence-level inference that no biochemical assay has tested.

Serine-protease inhibition by serpins requires an exposed reactive-centre loop that the target protease cleaves and is then trapped by. C28G1.2 retains a compact serpin-fold-like scaffold, but alignment to SRP-1, SRP-2, SRP-3 and SRP-6 shows no counterpart to their hinge and reactive-centre loop, and the gene is not among the five C. elegans serpins established as bona fide inhibitors. The activity is ancestral to the nematode clade (a PAINT IBD at the nematode node, seeded by SRP-2) but this lineage has lost the structural element the activity depends on. No C28G1.2 inhibition assay exists, so this is a sequence-level refutation.

The donor is RACK1, whereas direct worm studies establish WDR-23 substrate recruitment and SKN-1/GEN-1 regulation. WD-repeat architecture alone cannot transfer direct translational regulation. Human WDR23 ubiquitylates the histone-mRNA factor SLBP, but that is substrate-adaptor activity on a translation-related protein, not a translation regulator activity, and does not demonstrate this GO activity in S6FN32; an additional role is not experimentally excluded.

The weak sequence match to ribosomal RACK1 does not establish a ribosome-binding interface or ribosome association in this DCAF11/WDR-23 product. Direct WDR-23 studies support a different molecular system, but do not provide a negative ribosome-binding assay.