Horse benchmark review: pairs 31–40
All ten pairs have human research, GOA assessments, exact-sequence comparisons and ProtNLM claim assessments. UNDECIDED/UNC identifies unresolved evidence rather than an assumed biological error.
| Pair | Human annotation rows | Horse annotation rows | GO verdicts |
|---|---|---|---|
| SHLD2 | 26 | 0 | 4 UNC, 2 COR |
| GEMIN5 | 71 | 4 | 5 UNC, 1 LSP |
| AFAP1L2 | 18 | 1 | 9 COR |
| OLFML2A | 7 | 1 | 4 UNC |
| CC2D2A | 19 | 0 | 3 COR, 1 UNC |
| WDPCP | 26 | 2 | 4 COR |
| TRAF2 | 708 | 12 | 5 COR |
| WEE1 | 46 | 19 | 5 UNC, 2 COR |
| DYNLT2B | 32 | 0 | 10 UNC |
| DNMT3A | 87 | 26 | 1 CNN, 1 LSP |
Findings
- SHLD2: splice-product and domain evidence separates recruitment from DNA-end protection; the selected horse C terminus leaves repair claims uncertain. Human HPA actin localization is a real lead, not incompatible with nuclear function. The narrative is assessed separately.
- GEMIN5: a deletion at the end of the tandem WD40 region limits RNA-function transfer. Nuclear localization is less precise than the supported nucleoplasm annotation. Ubiquitination participation lacks a demonstrated mechanism.
- AFAP1L2: full-length conservation and primary adaptor experiments support the nine broad claims. The IL-6 result is preserved as an independently checked full-paper excerpt because the PMID cache is abstract-only.
- OLFML2A: extracellular family biology is supported, but the missing signal-peptide segment prevents certifying normal secretion and disulfide maturation of the selected protein.
- CC2D2A: N-terminal differences mainly affect the human disordered region, while the ciliary architecture is retained. Three developmental/pathway claims are supported; kidney-development specificity remains unresolved.
- WDPCP: primary structural data explain the human N-terminal extension. The conserved scaffold supports four broad claims; a separate deletion in the lipid-binding tail limits transfer of exact affinity.
- TRAF2: receptor binding, complex membership and immune processes are supported. Conflicting intrinsic-E3 experiments do not invalidate the strong cIAP-recruitment/ligase-complex evidence. Many human interaction-map rows remain unverified at partner level.
- WEE1: the protein is a full-length WEE1-family tyrosine kinase. WEE2-specific oocyte experiments suggest paralog transfer, but differential necessity and possible compensation do not definitively refute horse participation; all three oocyte/meiotic claims remain UNC.
- DYNLT2B: the C-terminal deletion removes a strand present in the experimentally solved light-chain fold. Exact-model function and localization remain uncertain despite high identity among the aligned residues.
- DNMT3A: the catalytic and ADD domains are retained despite a PWWP-end deletion. Broad nuclear gene regulation is supported; normal locus targeting is not assured. The protein-cysteine methylation assertion traces to a real in-vitro side reaction and is non-core. Horse testis literature supports gene-level expression/localization, not accession-specific function.
Validation and follow-up
The gene YAMLs undergo schema, GOA/reference and authored-term checks. Prediction sidecars undergo exact-source/title/excerpt validation, and the cohort inventory checks original output preservation. Gene pages are rendered, and curation history is scaffolded and validated separately from the YAML explanations. Core-function/deep-research advisory warnings do not justify adding AI prose as biological evidence.
Priority follow-ups are transcript/structure resolution of the altered horse models, primary assay/supplement tracing for human UNDECIDED rows, and horse oocyte evidence for WEE1. The initial review is complete as an assessment pass, not as a definitive resolution of every experimental annotation.