MitoMatch

What an AlphaFold-Multimer interactome of the mitochondrial proteome can and cannot do for GO curation

AI Gene Review · projects/MITO_INTERACTOME · 2026

Bottom line

  • Swaminathan et al. 2026 screened 630,003 mitochondrial protein pairs and report 2,895 predicted interactions (~85% precision), with partners for 85 of 101 orphans.
  • Predictions are hypotheses: never in existing_annotations, never a reason for protein binding.
  • The paper's experiments place orphan COA4 in copper delivery via COX11; we reviewed human COA4 and the yeast copper set (97 annotations).

From pairs to evidence

How the screen works

  • Score: mean ipTM over the five AFM models; averaging suppresses false positives better than the max.
  • Cutoff: first false positive near ipTM 0.5 on a copper-delivery benchmark (12 true pairs vs 805 non-interacting).
  • Driver: paired-MSA depth; over 97% of mitochondrial pairs clear it, which makes the organelle tractable.
  • Recovery: 56% of Complex Portal mitochondrial complexes fully recapitulated; TCAIM–OGDH and UQCC4–MT-CYB found blind.

COA4 joins the copper pathway

What the COA4 review looks like

Human COA4: complex IV assembly (IBA) accepted as core, now backed by the paper's knockout data. The BioPlex protein binding row is marked over-annotated.

Review results

Gene Species Ann. ACCEPT Non-core Over-ann. REMOVE
COA4 human 8 4 3 1 0
COA4 yeast 17 13 3 1 0
COX17 yeast 20 14 5 0 1
COX19 yeast 17 10 6 1 0
COX23 yeast 13 6 7 0 0
CMC2 yeast 12 6 5 1 0
PET191 yeast 10 6 4 0 0

The single REMOVE: COX17 protein farnesylation (GO:0018343), citing PMID:8078902, a paper about COX10. It seeded the miscitation audit.

Findings beyond the paper

  1. Symbols do not line up: there is no human COX23, and the paper's "PET191" is human COA5.
  2. GFP-library nucleus calls on COA4 and CMC2 flagged; cytoplasm calls accepted as a pre-import pool.
  3. Four accessory factors keep GO:0003674 (no MF): GO has no term for supporting a metallochaperone without binding metal.
  4. COX19 is a second COX11 chaperone, so COA4 and COX19 converge on the same target.

Status and next steps

  • ✅ Paper read and rules for using predicted PPIs written down.
  • ✅ Human COA4 and the yeast copper set reviewed; human COA5 already reviewed under the complex IV assembly module.
  • ⬜ Human COX17, COX19, CMC2; TCAIM, UQCC4; complex Q COQ3, COQ10A/B.
  • ⬜ Test species and conservation scores as a prior for IBA complex-membership rows.

Read more: projects/MITO_INTERACTOME.md · genes/human/COA4/ · genes/yeast/COX17/