Glycobiology

Auditing how GO annotates the enzymes and lectins of glycosylation, and mapping CAZy families to GO

AI Gene Review · projects/GLYCOBIOLOGY · reviewed 2026-10-04

Bottom line

  • We reviewed 376 annotations on 11 human glycogenes and indexed 17 pathway modules (100 more genes, 2,748 annotations).
  • Only 4 removals: the problem is altitude and pleiotropy (generic parent terms, downstream physiology), not wrong functions.
  • A CAZy→GO mapping (cazy2go) yields a 60-row safe propagation set and 34 hand-endorsed interpro2go gaps.

Why glycogenes

  • Glycosylation decorates most secreted and cell-surface proteins; defects cause >130 congenital disorders of glycosylation.
  • Glycosyltransferase families are large and sequence-similar, so propagated annotations can land on the wrong paralog or at the wrong level.
  • Specialist resources (CAZy, GlyGen, GlyConnect, GlyTouCan, GlycoCoO) hold knowledge GO may lack.
  • Question: where does GO over-annotate glycogenes, and where does it under-represent glycan biology?

Actions per gene

What the reviews found

  1. Generic MF → specific activity is the dominant fix: MGAT1 → GO:0003827, ST6GAL1 → GO:0003835, B4GALT1 → GO:0003831.
  2. The cellular-component version: bare membrane on Golgi transferases → GO:0000139 Golgi membrane.
  3. Guilt by substrate: PMM2 only supplies GDP-mannose, so N-linked glycosylation is non-core.
  4. Pleiotropy is not core: 62 of LGALS3's 106 rows kept non-core; lysophagy added as NEW.
  5. 8 new GO terms proposed across the seven exemplars.

Phase 3: mucin-type O-glycans

The module cohort

One of the 17 indexed glycobiology modules (N-glycan LLO lumenal assembly). The 100-gene module cohort shows the same skew: 0.8% REMOVE, 63.3% ACCEPT, 17.6% non-core, 14.0% over-annotated.

cazy2go: CAZy families → GO MF

Step Result (2026-06 snapshot)
Families with a GO MF via member ECs 283
Fully masked by interpro2go 56 (20%)
Safe propagation set after filters and hand review 60 rows
Hand-reviewed true gaps 34 ENDORSE · 13 CAUTION · 6 REJECT
Poly-specific families resolved to subfamily signatures 66 of 90

Rejects were CBM (non-catalytic) families and one GT family carrying a protease EC from another domain; both are now filtered automatically.

Status and next steps

  • Done: 11 exemplar reviews; 17 modules indexed; cazy2go safe set built.
  • Next: run the GOA closure query to size the animal glycogene set and its baseline.
  • Next: submit the 8 proposed terms and the 34 endorsed interpro2go gaps after curator sign-off.
  • Next: remaining GALNT paralogues, core 3/4 and capping steps; re-review modules for new terms.

Read more: projects/GLYCOBIOLOGY.md · projects/GLYCOBIOLOGY/