Metabolic model analysis

Using genome-scale metabolic models to catch GO molecular-function errors

AI Gene Review · projects/METABOLIC_MODEL_ANALYSIS · 2026

Bottom line

  • Compared EC numbers in 3 GEMs (M. extorquens iRP911, E. coli iML1515, human Recon3D) with UniProt: 42 / 51 / 61% of aligned genes disagree.
  • Much is expected (EC class 7, complex-subunit EC propagation), but 8 gene reviews found real errors on both sides.
  • GO wrong: mdcD (7 rows removed), ecm, HADHB. Model wrong: rbsD, glgX, HADHB, CPT1C. FBA shows the errors matter.

How often models and UniProt disagree

Not every disagreement is an error

  • EC class 7 (translocases) was created in 2018; iRP911 is from 2011. Cytochrome c oxidase 1.9.3.1 → 7.1.1.9 is a reclassification.
  • Complex convention: a GPR lists every subunit, so every EC of the complex lands on each gene. E. coli lpd picks up 10 ECs.
  • So GEM gene→EC pairs must be filtered before they can validate GO MF terms.

When UniProt and GO were wrong

Gene (organism) Model says GO/UniProt said Review
mdcD (M. extorquens) 4.1.1.89 decarboxylase 6.4.1.3 carboxylase (ligase) 7 REMOVE; carboxy-lyase ACCEPT; malonate catabolism NEW
ecm (M. extorquens) noEC (2011) methylmalonyl-CoA mutase MODIFY to GO:0016866 intramolecular transferase; no GO term for EC 5.4.99.63 yet
HADHB (human) — hydratase, dehydrogenase (TAS) REMOVE: those are HADHA's

mdcD shares 35% identity with an acetyl-CoA carboxylase subunit; one wrong EC cascaded into GO, keywords and family assignments.

When the model was wrong

FBA: do the errors matter?

Test Original model Corrected / reality
E. coli rbsD KO on ribose 100% growth (rbsD in transporter GPR) 0% after adding the pyranase step
Recon3D HADHA KO, FAOXC5C3x flux 1000 → 0 essential, as expected
Recon3D HADHB KO blocks only HISDr (histidase) should block thiolase steps

E. coli glgX was also wrong in iML1515: a debranching enzyme modelled as branching enzyme (2.4.1.18 vs 3.2.1.196).

Status and next steps

  • Reviews: genes/METEA/{ecm,sucB,mdcD,gcvP}, genes/ECOLI/{rbsD,glgX}, genes/human/{HADHB,CPT1C}.
  • Open: M. extorquens mdcB, ilvC, purK; remaining Recon3D CLASS_CHANGE genes.
  • Caveat: the models/metabolic/ files and FBA scripts cited on the page are not in this repository.

Read more: projects/METABOLIC_MODEL_ANALYSIS.md