ACTL10

Summary: Phylogenetic inference placing ACTL10 in the actin cytoskeleton, from PANTHER node PTN002631484. This row is byte-identical to the corresponding row in the already-merged ACTL8 review - the same node and the same twenty-five WITH/FROM tokens - and it is resolved the same way, for the same reason, which is a property of the node rather than of either gene. PTN002631484 spans the conventional actins, the POTE actin-fusion genes and the divergent actin-like proteins including ACTL8, ACTL9, ACTL10 and ACTRT1-3, so identity to beta-actin across its human descendants runs from about a third to 100 per cent. The term it donates is correspondingly generic. That breadth is therefore not a granularity failure: the term is the true last common ancestor of a heterogeneous donor set, and GRANULARITY_MISMATCH would be the wrong call because the donors do not agree on anything more specific. No specificity upgrade is warranted. Verified independently rather than read off the sibling review: QuickGO confirms PTN002631484 donates this term to exactly eighteen human genes - ACTA1, ACTA2, ACTB, ACTBL2, ACTC1, ACTG1, ACTG2, ACTL8, ACTL9, ACTL10, ACTRT1, ACTRT2, ACTRT3 and the five POTE genes. So this row is one projection across eighteen genes with identical evidence, not an independent finding about ACTL10, and the same is true of the GO:0005200 row below, which shares this row's reference GO_REF:0000033. That is the correct reading of a family-level inference and it is why the row is non-core. One thing this check ruled out. The merged ACTL8 review's central finding is that PANTHER mis-placed ACTL8 inside the narrow beta/gamma-actin subfamily, inflating its record about five-fold. Querying QuickGO directly for all eight divergent human actin-like and actin-related-T proteins confirms that finding and shows it does not extend to ACTL10: ACTL8 alone sits under the narrow nodes PTN002631586 and PTN007551913 and carries eleven IBA rows, against a median of two for the other seven, and ACTL10 has exactly two and sits under neither narrow node. ACTL10 is therefore correctly placed; its problem is not mis-placement. The donor set is strong rather than weak. Of twenty-five WITH/FROM tokens, twenty-four are protein identifiers, all twenty-four resolve, and all twenty-four carry their own experimental-code annotation for this term or a descendant, spanning twelve organisms. Two resolve only to unreviewed TrEMBL entries (Candida ACT1 and C. elegans act-5); that is recorded rather than hidden, and both still carry their own IDA, so their weakness is one of name provenance, not evidence provenance. It is kept because it asserts nothing that ACTL10's sequence contradicts - ACTL10 is a genuine divergent actin, about 35 per cent identical to beta-actin, and on the extended reading frame actin's phosphate-binding loop 1 is conserved, though the reviewed 245-residue sequence stops short of it - and it is marked non-core because nothing places ACTL10 in any compartment experimentally. Unlike ACTL8, whose UniProt entry at least offers a by-similarity cytoplasm/cytoskeleton location, ACTL10's entry carries no SUBCELLULAR LOCATION line at all; its only CC line in the entire record is a bare family-membership statement with evidence code ECO:0000305, which is curator inference rather than similarity to a named entry. A generic compartment term cannot represent the core of a gene whose activity is unknown. One qualifier point, raised because it asserts more than the evidence does. This row carries is_active_in, which states that ACTL10 carries out its molecular function in the actin cytoskeleton. No molecular function has been measured for ACTL10 at all, so is_active_in presupposes exactly what is unknown; located_in would say what the inference actually supports, namely residence, without committing to activity. The donors do justify is_active_in for themselves, since conventional actins demonstrably act in the actin cytoskeleton, so this is a transfer artefact of the same kind as the term's breadth rather than a curator error. The term is left unchanged and the action stays KEEP_AS_NON_CORE, because the compartment call is defensible and swapping a qualifier is a GOA-side change rather than a term-level one; it is recorded here and raised for PAINT as the narrower of the two recommendations on this gene.

Reason: Correctly scoped family-level inference from a node at the right depth for a divergent actin, with every protein donor carrying its own experimental evidence. Retained because the sequence is consistent with it, marked non-core because no experiment has localised ACTL10 anywhere and UniProt offers not even a by-similarity location. Resolved identically to the byte-identical row in the merged ACTL8 review, so the two reviews agree.

Summary: Phylogenetic inference of a specific molecular function from PANTHER node PTN000940351, on a protein for which no molecular function of any kind has been measured. This row is byte-identical to the corresponding row in the merged ACTR10 review, which accepted it as core; it is resolved differently here, and the ground of the difference is gene-specific evidence rather than a difference of method. Why ACTR10's ACCEPT does not transfer: ACTR10 is a bona fide dynactin subunit, and its accepted reason is that its own contribution to dynactin's structural integrity is exactly what this molecular function expresses. ACTL10 has no known complex, no known location, no phenotype and no biochemistry, so there is no corresponding gene-specific basis. Why the REMOVE that ACTL7A and ACTL7B applied to their GO:0005200 rows also does not transfer: both of those rested on two grounds that are absent here. Theirs was a TAS row citing a paper that contained no functional experiment, whereas this is a live IBA whose ten protein donors all carry their own experimental evidence; and PAINT had explicitly negated the term at their node PTN008986528, whereas PAINT has not negated it on ACTL10's branch. What the donor analysis does and does not license. All eleven WITH/FROM tokens were resolved: ten are proteins, ten of ten resolve, and ten of ten carry their own experimental-code annotation for the term - mouse and rat Actg1 by IDA, yeast ACT1 and ARP1 by IDA, yeast ARP10 by IPI, human ACTB by EXP, IDA and IMP, human ACTR2 and ACTR3 by IDA, and two Dictyostelium actins by IDA. So SOURCE_WEAK_OR_INFERRED and SOURCE_EVIDENCE_WEAK are contradicted by the analysis itself; the defect is not in the sources but in the transfer, hence PROPAGATION_BAD. Two further checks, both of which sharpen the call. First, this row is also a many-genes-one-source projection: QuickGO shows PTN000940351 donating the term to exactly ten human genes with identical evidence, and the row shares its reference GO_REF:0000033 with the GO:0015629 row above, so the two rows are not independent of each other either - six of these ten protein donors also appear on that row. Second, and more telling, the four donors that are unique to this row, and therefore the only ones that make it more specific than a generic actin-cytoskeleton call, are yeast Arp1, yeast Arp10, human Arp2 and human Arp3. Arp2 (P61160) and Arp3 (P61158) are seeds at PTN000233596 and PTN000233796 respectively, and those are two of the eight nodes at which PAINT has IRD-negated this very term; yeast Arp1 is a dynactin filament subunit, and the centractin clade node PTN007551901 also carries an IRD negation. So the specificity of this row rests largely on donors drawn from clades that GO's own pipeline has excluded from the term they are donating. The substantive argument is structural, and it was rebuilt on a corrected sequence. The route by which real actins earn this term is filament formation. Measured on PDB 6DJO, the 38 inter-protomer contact positions are 38 of 38 chemically compatible in beta-actin and alpha-skeletal actin, and 33 of 38 in Drosophila Arp53D, a divergent actin that does polymerise; ACTL10 scores 11 of 38, the same as ACTL8 and the same as the independently annotated Sapajus ACTL10. That is a real degradation of the polymerisation surface and it is stable under a second substitution matrix and gap model. REMOVE is nevertheless not earned, and the control that refuses it is one this review computed itself: human Arp3 scores 8 of 38, below ACTL10, and Arp3 does form actin-like protomer contacts as the first protomer pair of a daughter filament at an Arp2/3 branch. A low score on this metric therefore bounds canonical incorporation into a conventional two-stranded filament; it does not show that a protein cannot be a structural component of any actin-containing assembly. The complementary evidence points the other way as well: the nucleotide pocket is well preserved on the extended reading frame, 15 of 19 contact positions chemically compatible with a fully intact phosphate-binding loop 1, which is better than ACTL8 and better than ACTL7A. On the reviewed 245-residue sequence the loop is not present to score at all, which is the point of the correction rather than a competing measurement. So this is not a fold that has lost actin's residues, and no argument from residue decay is available. The term is marked over-annotated rather than removed: it is not clearly false, but it asserts a specific molecular function for a protein with no functional data, propagated from a node whose term GO's own phylogenetic pipeline has judged non-transferable to eight other divergent-actin clades, dropping to the parent GO:0005198 where it wanted to retain anything at all. Why it is not MODIFY to that parent, which is what the merged ACTRT3 review did with this same row. ACTRT3's generalisation is well founded, and explicitly so: its reason states that GO:0005198 is supported for ACTRT3 because it is a component of the perinuclear theca's protein scaffold and contributes to its structural integrity, so generalising moves the row from an unsupported specific claim onto a supported general one. ACTL10 has no counterpart to that. It is in no known complex or assembly, has no known location, no phenotype and no binding partner, and GO:0005198 is defined by contribution to the structural integrity of a complex or assembly - so substituting it would exchange one unsupported assertion for a slightly less specific unsupported assertion, and would newly place a molecular function on a gene for which none is supported - ACTL10's only existing MF row is the GO:0005200 under discussion here. Neither ACTL10 nor ACTRT3 already carries GO:0005198, so for ACTRT3 the MODIFY adds a defensible term and for ACTL10 it would add an undefensible one. Flagging the existing row as an over-annotation says exactly as much as the evidence supports and no more. This is the same kind of gene-specific discrimination ACTRT3 itself uses to distinguish its verdict from ACTR10's ACCEPT.

Reason: A specific molecular function asserted on a protein with no measured function of any kind. The polymerisation surface through which actins earn this term is substantially degraded in ACTL10 (11 of 38 interface positions compatible, against 38 of 38 for beta-actin and 33 of 38 for a divergent actin that does polymerise), and PAINT has already negated this exact term at eight sibling divergent-actin nodes, including the clades of three of the four donors that are unique to this row. REMOVE is withheld because the review's own control declines to support it: Arp3 scores lower still and does make actin-like protomer contacts, so the metric bounds canonical filament incorporation rather than all structural roles, and the nucleotide site is well preserved rather than decayed. Every donor annotation is itself sound, so the failure is in the transfer rather than in the sources.

Summary: An inter-ontology inference whose literal WITH/FROM field is not a gene product but the GO term GO:0005200, i.e. it is derived entirely from the molecular-function row above and carries no evidence independent of it. It is therefore exactly as strong as that row and no stronger, and it is marked at the same level. The two merged siblings that share this byte-identical row diverge, and both routes are unavailable here. ACTL7A modified it to the more specific GO:0030036 actin cytoskeleton organization, but that was earned by a knock-out phenotype in which Actl7a-null spermatids lose subacrosomal F-actin; ACTL10 has no phenotype to support any process term, and the one paper to look at mouse Actl10 in a germline context states that its mutants have not been linked to germline biology. ACTL7B removed it outright, but only because the GO:0005200 row it derives from was itself removed; here that row is marked over-annotated rather than removed, so removing the derived row would assert more than the review does about its basis. ACTR10 kept it as non-core on the grounds that it is true but uninformative - which is defensible for a gene with a demonstrated cytoskeletal role, and is precisely what ACTL10 lacks. ACTRT3, which merged with the byte-identical row, also kept it as non-core, and its reason is the one that shows why ACTL10 cannot follow: it holds that the term "is true at this level of generality and should not be removed" because ACTRT3 is a perinuclear-theca scaffold component required for acrosome assembly. Truth at that level of generality is exactly what is unestablished for ACTL10, which has no demonstrated involvement in organising anything. So for ACTR10 and ACTRT3 the row is true but uninformative, whereas for ACTL10 it is unsupported - a difference that MARK_AS_OVER_ANNOTATED records and KEEP_AS_NON_CORE would hide. Marked over-annotated rather than kept: read as an assertion that ACTL10 participates in organising the cytoskeleton, this row overstates what is known, because its entire basis is a molecular-function inference that this review has judged an over-annotation. The coded diagnosis records the circularity rather than weak source evidence, matching the merged ACTR10 and ACTRT3 reviews, which both give this source CIRCULAR_OR_REDUNDANT. The defect here is not that the source's evidence is weak - the source is not a gene product at all and carries no evidence to weigh - but that the chain transfers from another transfer on the same gene. SOURCE_EVIDENCE_WEAK would also contradict this review's own finding on the row above, where every donor was shown to carry its own experimental annotation. ACTL7A and ACTL7B use the weak-source values instead, which is right for them because their source row is a 1999 TAS with genuinely weak evidence, and wrong here.

Reason: Wholly derivative of the GO:0005200 row, whose literal WITH/FROM it is, and therefore no stronger than that row, which is itself marked as an over-annotation. No independent evidence, no phenotype and no localisation support a cytoskeleton-organisation role for ACTL10, so the annotation is retained for traceability but should not be read as evidence that ACTL10 organises anything. It is not modified to the more specific GO:0030036, because the phenotype that earned that term for ACTL7A has no counterpart in ACTL10.