Biosynthetic gene cluster complexes

Reviewing obligate enzyme pairs from natural-product clusters

AI Gene Review · projects/BGC · 2026

Bottom line

  • Many BGC enzymes work only as heteromeric pairs, and GO often gives the pair's activity to the wrong subunit.
  • Using an AlphaFold3 screen of 2,437 MIBiG clusters as structural evidence, we reviewed 3 of 5 PDB-backed exemplar pairs (6 genes, 29 rows).
  • Two reusable patterns: non-catalytic partners inherit their fold's catalytic term, and a heme-less P450 pseudoenzyme inherited a full P450 cofactor set (all removed).

Why complexes, and the evidence used

  • In a cluster, the product constrains each enzyme's function: a strong prior for checking annotations.
  • Moriwaki et al. (bioRxiv 2025, v2 2026): AlphaFold3 with MMseqs2 over 487,828 protein pairs; 15,438 predicted heteromers at ipTM ≥ 0.6, 381 high-confidence structurally homologous pairs.
  • Exemplars come from the paper's PDB-backed validation set, so each prediction matches a solved structure.
  • Predictions are hypotheses: solved complexes can score low (PDB 7YN3 at ipTM 0.17).

Three pairs, one failure mode

A pseudoenzyme stripped of P450 terms

eryCII review: UniProt says it "lacks the heme-binding sites" and the 2YJN structure has no heme. Monooxygenase, heme, iron and oxidoreductase rows removed; GO:0008047 enzyme activator activity added.

Results

Gene Role Rows Changes
pqsB non-catalytic partner 4 acyltransferase over-annotated
pqsC catalytic 7 fatty acid BP removed; KAS III MF modified
actI-ORF1 KSα, catalytic 5 fatty-acid terms modified or removed
actI-ORF2 CLF, no active site 2 both acyltransferase rows over-annotated
eryCIII glycosyltransferase 6 UDP-GT removed (donor is TDP-sugar)
eryCII pseudoenzyme activator 5 4 P450 rows removed; activator NEW

29 rows: 15 ACCEPT, 7 REMOVE, 3 MODIFY, 3 MARK_AS_OVER_ANNOTATED, 1 NEW. Proposed terms: 2-heptyl-4(1H)-quinolone synthase activity; polyketide chain length factor activity.

Status and next steps

  • ✅ PqsBC, actinorhodin KS-CLF, EryCIII/EryCII reviewed; erythromycin cluster captured as a pathway concept (terms/erythromycin_biosynthesis/).
  • ⬜ Nosiheptide (RiPP) and pyoluteorin (low-ipTM control) pairs queued.
  • ⬜ Novel, non-validation predictions go to -predictions-review.yaml files.
  • ⬜ The status table on the project page still lists the ActVA and DEBS rows, which share PDB 1TQY and 2YJN with the reviewed pairs.

Read more: projects/BGC.md · genes/PSEAE/pqsB/ · genes/STRCO/actI-ORF2/ · genes/SACEN/eryCII/