PDB structures as GO evidence

Where deposited structures change a gene's annotation, and where they only confirm it

AI Gene Review · projects/PDB · 2026

Bottom line

  • 949 of 2,529 pipeline genes have a deposited structure (13,415 PDB entries), yet only 15% of 737 structure-paper and gene pairs are cited in GOA.
  • Structures reliably give under-curated proteins their first experimental-grade evidence (~12 annotations, 4 NEW rows in round 2).
  • New informative function is rare: 2 structure-unique annotations (both merA); the papers' own hypotheses paid off clearly in 1 of 6.

Why look at structures

  • A bound cofactor, metal, ligand or partner is direct evidence for MF, binding and complex membership.
  • Rank candidates by richness × sparsity: a cofactor-bound, full-length structure of an IEA-only enzyme scores high.
  • Three cuts (278 genes): dark_mf 114 · euk 100 · contested 130 (a review disputed a catalytic MF).

Structure papers are missing from the evidence trail

174 of 247 genes cite none of their structure papers. GAP_OPPORTUNITY: the paper predates the gene's latest experimental annotation.

Does structure fill gaps? (H1)

Structure adjudicates contested activities

Gene Disputed MF Action In the structure
HEN1 (ARATH) peptidyl-prolyl isomerase REMOVE SAH → methyltransferase
XYL1 (PICST) oxidoreductase (generic) REMOVE NADP
DOT1 (yeast) methyltransferase (generic) over-annotated SAM/SAH
SIRT2 (human) transferase (generic) over-annotated NAD, Zn
mcrA (METAC) transferase (generic) REMOVE F430, SAM, Fe-S

Two cases: an over-general parent (cofactor confirms catalysis, points to the specific child) versus a wrong specific activity (structure argues against it).

Caveats we hit

  • RCSB per-entry citations drift: SPR's are inhibitor screens, cbh1's glycosylation NMR. Verify each PMID before citing.
  • A bound ligand is a clue, not proof; additives are filtered but adventitious binders remain.
  • Round 1 tested genes that were already curated experimentally, so structures could only corroborate (4 of 4).

Status and next steps

  • ✅ Inventory, RCSB enrichment, citation-gap and H1 ledger
  • ✅ Structure-informed reviews: XYL1, IDH3B, COX6B1, psaC, COI1, ATAD1; merA, mcrA, secA, mxaI
  • ⬜ Work down GAP_WORKLIST.md (PNO1, RPS3, BIRC5, GCH1, SIRT2 …)
  • ⬜ Map complex partners to UniProt; consider a PDB-evidence field in the schema

Read more: projects/PDB.md · projects/PDB/H1_LEDGER.md · projects/PDB/CURATION_GAP.md