Integrated the FutureHouse Falcon deep-research report (B3GALNT2-deep-research-falcon.md) into the review. The report's conclusions broadly agree with the existing review (core M3 / matriglycan biology, ER-primary localization, avoidance of apoptosis/inflammation/pyroptosis/synaptic over-annotations). Changes made:
reference_review: relevance MEDIUM, correctness VERIFIED. This is the one genuinely new primary paper in the report.supported_by on the IBA Golgi-membrane annotation to back the KEEP_AS_NON_CORE call.description to note the N-glycan/intracellular type-I LacdiNAc activity (kept project-independent, flagged its significance as not yet established) and to state ER-mainly/Golgi-partly localization.suggested_question to reflect that the N-glycan substrate question is now partly addressed by PMID:30898876.Falcon claims NOT acted on (with reasons):
- All other report citations are review/contextual papers (Praissman & Wells 2014, Sheikh 2017, Willer 2014, Endo 2015, Bouchet-Séraphin 2015, Bigotti & Brancaccio 2021, Sharaf-Eldin 2025, Togayachi 2026). They restate known core M3 / matriglycan pathway and disease biology already captured by the existing references and primary citations (PMID:23929950, PMID:23453667, PMID:14724282, Reactome). Adding them as citations would not strengthen any specific annotation; not added.
- Falcon repeatedly frames B3GALNT2 localization as "ER, not primarily Golgi" and at one point implies the Golgi annotation should be downweighted further. The existing review already keeps Golgi as non-core; the Nakane primary data ("mainly ER, partly Golgi") actually justifies retaining the Golgi annotation rather than removing it, so no REMOVE was applied (consistent with not overruling on partial evidence).
- Falcon's muscle/brain/ECM "biological process" discussion describes downstream disease consequences (cobblestone lissencephaly, sarcolemmal integrity, laminin binding). These are organism-level phenotypes of the glycosylation defect, not direct B3GALNT2 GO process functions; no new BP annotations proposed (the review already centers on GO:0035269 O-mannosylation). Consistent with Falcon's own annotation-risk assessment.
- No PMID was added for any claim that could not be resolved/fetched; the only resolvable new primary paper was Nakane 2019.
This entry supersedes the earlier family-default localization, independent-imaging, ontology-gap and generic-binding-overannotation interpretations above. The earlier journal and Falcon report remain unchanged as historical records. The present evidence supports one GalNAc-transfer activity in the endoplasmic reticulum, with direct extension of the alpha-dystroglycan core M3 O-mannosyl glycan and qualified cultured-cell N-glycan substrate breadth. B3GALNT2 performs the transfer step; POMK performs subsequent phosphorylation. The glycopeptide binding experiment in PMID:23929950 does not assay B3GALNT2 protein binding to laminin, so the earlier categorical enzyme-binding statement should not be treated as an experimentally established negative.
The 16 source annotations and both alternative products are preserved. Decisions are 6 ACCEPT, 2 KEEP_AS_NON_CORE, 7 MODIFY and 1 UNDECIDED, with no new annotation. The general electronic glycoprotein-biosynthesis assertion is retained to cover supported N/O substrate breadth; the source-specific patient-study assertion is refined to the O-mannosyl pathway with the direct core M3 enzyme work as additional support. Broad membrane localization is refined to ER membrane rather than marked excessive. The catalytic evidence comes from soluble engineered constructs and cannot automatically be assigned equally to the substantially altered natural isoform Q8NCR0-2.
The PAINT Golgi assertion is a curated ancestral inference at PTN000130249, not an automatic family default. The O-glycosylation and UDP-glycosyltransferase inferences at PTN002257389 legitimately use experimental Q8NCR0 evidence; this self-grounding is not circular. The local node records were inspected, but a complete family tree and alignment were not. Golgi remains non-core with explicit curator deference while the core summary uses only ER membrane.
PMID:14724282 was read as a complete normal abstract: human soluble catalytic-region assays and NMR establish the beta-1,3 GalNAc linkage, without establishing all tested acceptors as major physiological substrates. PMID:23929950 was read through the complete cached experimental narrative; supplementary Methods, raw spectra and figure pixels were not inspected. PMID:23453667 and PMID:30898876 carry true full-text flags but their normal caches omit Results and Methods. Selected indexed original localization Methods/Results/Figure 4 for the former and selected N-glycan Results/Figure captions 1/5/6/8 plus cell-culture/transfection Methods for the latter were read separately. No complete supplementary datasets or figure pixels were inspected.
Human wild-type B3GALNT2 constructs colocalize with ERp72 in mouse C2C12 cells in PMID:23453667. The ER/partial-Golgi discussion sentence in PMID:30898876 cites that study; it is not a second independent imaging result. The latter study includes endogenous HeLa knockdown readouts for WFA-captured NCSTN/LRP1 and separate overexpression/Lec8 hamster experiments. WFA specificity and overexpression constrain linkage/substrate interpretation; endogenous tissue functions remain incompletely resolved. The electronic glycoprotein-biosynthesis annotation is not restricted solely to the dystroglycan acceptor on these data.
For PMID:32296183, the complete abstract and selected main screening, validation and limitation paragraphs were read. TMBIM1/Q969X1 is the known GOA WITH/FROM partner and appears in immutable UniProt. The exact original target experiment, ORF/isoform and pair-specific validation could not be inspected because IntAct access failed. This source-specific uncertainty supports UNDECIDED, not a finding that the interaction is false, excessive or unrelated simply because the term is generic. No more specific binding function is invented. Three curated records are not claimed to represent three independent publications.
Both human Reactome summaries were read completely. Official indexed GO definitions support acetylgalactosaminyl transfer and glycan elongation within the O-mannose process. The bounded ontology search did not prove absence of a more specific term for EC 2.4.1.313/RHEA:37667; the categorical new-term claim is therefore withdrawn and retained as a question. No local GO-CAM index hit was found, which is not evidence of a missing annotation or of global model absence. The new short YAML anchors are exact normal-cache substrings and total at most 25 words per source; historical journal quotations above were preserved, not reissued as new evidence.
Focused validation completed successfully on 2026-09-29 with four advisories and no blocking errors. Two advisories concern absent structured propagation-review objects for the refined IBA terms; the reasons retain the actual ancestral nodes and legitimate target-derived experimental grounding, without claiming a complete tree/alignment audit or a propagation failure. The glycoprotein-biosynthesis action difference is intentional: the general electronic assertion retains supported N/O substrate breadth, while the patient-study assertion is refined to its supported O-mannosyl context. The preserved Falcon report is not cited as evidence because its superseded interpretations are replaced by primary-source findings. HTML rendering passed. These are focused gene checks; no repository-wide validation success is claimed.
This entry supersedes the preceding decision to leave the reaction-specific ontology issue only as an expert question. The underlying chemical distinction is well established: B3GALNT2 transfers GalNAc onto the terminal GlcNAc of the O-mannosyl glycan, whereas initiating polypeptide GalNAc transfer to Ser/Thr is a different reaction. A candidate ontology request is restored for curator assessment, without claiming that a complete current GO search proved absence or assigning a new GO identifier. The official IUBMB entry defines EC2.4.1.313; the complete returned AmiGO GO:0008376 immediate-child list contained no matching core-M3 reaction, but its displayed load date was 2026-08-06. Indexed queries do not exclude an unexpected parent or equivalent existing term, and direct OLS/QuickGO descendant access was unavailable. The broad GO:0008376 annotation remains until an appropriate refinement is established.
The core evidence now places the original NMR linkage excerpt (PMID:14724282), ER-localization excerpt (PMID:23453667), and trisaccharide/POMK-recognition excerpts (PMID:23929950) together. The corresponding reference-finding quotes were redistributed rather than expanded without limit; all current YAML excerpts remain exact normal-cache substrings and total at most 25 words per source. The first source is abstract-only. The ER quote itself is from the abstract; the separately read original localization experiment uses human constructs in mouse C2C12 cells. The POMK-recognition wording is from the closing Discussion and supports substrate ordering, not B3GALNT2 kinase activity. No new complete-paper, supplemental-data or figure-pixel reading is claimed.
Both Golgi annotations remain non-core on positive evidence: PAINT provides the ancestral inference, while UniProt records Golgi membrane by similarity and supplies the SL-0134 mapping. The citation-dependent sentence in PMID:30898876 remains distinct from independent localization imaging. Canonical type-II topology and EC2.4.1.313 are restored in the biological description; UniProt topology features are sequence based and the type-II assignment uses similarity. They do not establish identical topology or catalytic function for the substantially altered Q8NCR0-2 product. The core donor CHEBI:67138 uses the official charged-species label UDP-N-acetyl-alpha-D-galactosamine(2-), independently checked on ChEBI and linked in the immutable UniProt catalytic reaction.
The PMID:32296183 reference is marked VERIFIED for independently checked PubMed identity and its HuRI study context. This does not resolve the original B3GALNT2–TMBIM1 experiment or constructs, so the IPI annotation remains UNDECIDED under the user-defined evidence-access rule. No generic-binding removal or new specific interaction function is inferred. All 16 source assertions, all 16 actions (6 ACCEPT, 2 KEEP_AS_NON_CORE, 7 MODIFY, 1 UNDECIDED), both alternative products and the single core activity/process/location are preserved. No new biological-process annotation is proposed; a candidate ontology term is separate from a NEW gene annotation.
Focused validation after this follow-up passed with four advisories and no blocking errors. The two structured IBA-metadata cautions retain the recorded node evidence without claiming a complete tree/alignment review. The glycoprotein-biosynthesis action difference remains source-specific, and the preserved Falcon report is not cited as authoritative evidence for the revised interpretation. Rendering passed. All 16 source assertions, their actions, both alternative products and downloaded sources remain unchanged; no repository-wide validation success is asserted.
The current-head review of PR #3563 (commit 00c22ed89, submitted 2026-09-30) identifies a policy disagreement: its requested REMOVE for GO:0005515 applies the default informational-exclusion criterion. That criterion can exclude a generic term even when the reported interaction is real; it is distinct from concluding that the experimental evidence is false. This concern is acknowledged rather than recast as a biological objection.
For this task, the explicit evidence-access requirement is to use UNDECIDED when the relevant experiment cannot be adjudicated. The B3GALNT2–TMBIM1 pair and primary HuRI citation are identified, but the target experiment, tested ORF/isoform and pair-specific validation have not been inspected. Reading general HuRI screening methods and observing full_text_available: true do not establish that those target-level materials were accessed. The bibliographic VERIFIED flag and the annotation's UNDECIDED action therefore describe different judgments.
The task also directs that generic binding not be changed solely for informativeness. That instruction does not establish the pair's function or supersede the evidence-access requirement with blanket retention: UNDECIDED remains the present action. No more specific molecular function is inferred, no global skill policy is changed, and all 16 actions, both products, the core function, candidate term and reference findings remain unchanged. Revisiting the action requires actual target-level evidence and application of the instructions governing that review, not merely the cache availability flag.