Protein complex functions

Which subunits should carry a complex's molecular function?

AI Gene Review · projects/PROTEIN_COMPLEX_FUNCTIONS · 2026

Bottom line

  • Complex activities get assigned to every subunit, turning membership into a false catalytic function.
  • We wrote a decision framework and backed a GO-CAM split into active vs accessory/structural members, so only active members export MF.
  • Tested on the proteasome (PSMB5 vs PSMA1) and BGC heterodimers; structure prediction gave no curation-grade interface.

Same complex, different roles

Working principles

  1. Complex membership alone is not evidence for a molecular function.
  2. Essential for an activity ≠ the executor of it.
  3. contributes_to only for direct participants: catalytic core, electron relay, substrate binding, mechanical coupling.
  4. Assembly factors get assembly terms unless they stay in the active complex.
  5. Multi-activity complexes (ribosome, proteasome): assign roles per activity.

A catalytic subunit, reviewed

PSMB5 (β5): threonine-type endopeptidase activity ACCEPTed; generic endopeptidase MODIFY to the specific term. Its alpha-ring partner PSMA1 carries no peptidase rows.

Worked examples from BGC heterodimers

Complex Catalytic member (enables) Partner's role
PqsBC PqsC, acyltransferase PqsB: contributes_to
Act KS-CLF ActI-ORF1, polyketide synthase CLF: chain-length factor
EryCII-EryCIII EryCIII, glycosyltransferase EryCII: enzyme activator activity

All three had the catalytic MF on the non-catalytic partner in GOA via domain-signature propagation.

Can structure prediction assign roles?

Status and next steps

  • Done: framework, rubric, GO-CAM export position, PSMA1/PSMB5 reviews, Boltz2 and ESMFold2 pilots.
  • Open: OXPHOS attribution matrix; audit OXPHOS reviews for contributes_to and assembly-factor consistency; guidance for enrichment and ML-label users.

Read more: projects/PROTEIN_COMPLEX_FUNCTIONS.md · projects/OXPHOS.md · projects/PROTEIN_COMPLEX_FUNCTIONS/esmfold2/