Top-Nots

Mining ~6,300 reviewer REMOVE decisions for candidate NOT annotations

AI Gene Review · projects/TOP_NOTS · 2026

Bottom line

  • A NOT annotation is the only way to stop pipelines re-asserting a wrong activity; a REMOVE cleans one review.
  • A keyword score over ~6,319 REMOVE/over-annotated rows found 250 candidates in 19 species, 115 strong, 22 Tier 1.
  • None has yet been literature-verified or filed as a NOT: this is a worklist, not a submission.

Why a REMOVE is not enough

How candidates were found

  • Scan every *-ai-review.yaml for REMOVE or MARK_AS_OVER_ANNOTATED.
  • Score the review summary for negative-evidence phrases: "does not have", "lacks catalytic", "pseudoenzyme", "no detectable activity", "does not catalyze", "no intrinsic".
  • ≥ 6 very high confidence · 4–5 strong · 3 moderate, manual review.
  • A keyword hit is a lead: a formal NOT needs positive evidence of absence.

The funnel

Tier 1 examples

Gene Wrong term Why it is absent
DROME CG6051 PI3P phosphatase (IEA) FYVE protein: binds PI3P, has no phosphatase activity
ANOGA PGRPS1 amidase (IBA) lacks the zinc-binding residues
human AKTIP ubiquitin-conjugating enzyme (IBA) UEV domain lacks catalytic cysteine
human PLD5 catalytic activity (IEA) lacks conserved HKD motifs
yeast YDJ1 ATP binding (IEA) DnaJ stimulates Hsp70 ATPase, binds no ATP
human SURF1 cytochrome-c oxidase (IEA) assembly factor, not the enzyme
human RASA1 GTPase activity (TAS) GAP: stimulates Ras, no intrinsic GTPase

Patterns

  1. Pseudo-enzymes (highest value): Epe1, PLD5, AKTIP, AIP, CG6051, CPT1C, Pld4.
  2. Domain ≠ function: dockerins as hydrolases; DnaJ domains as ATP-binding (YDJ1, DNAJA2, DNAJA4, Dnaja3, DnaJ).
  3. "Is phosphorylated" ≠ "does phosphorylation": ~19 human genes on GO:0006468. Largest by count, but mostly removals, not NOTs.
  4. Assembly factor ≠ complex activity: SURF1, ATP10, IDH3B, Ccs.
  5. Upstream regulator ≠ direct activity: EGFR as MAP3K; RASA1 as GTPase.

Status and next steps

  • ✅ Scan, scoring, pattern categories; 250 candidates ranked.
  • ⬜ Verify the literature behind each Tier 1 negative claim.
  • ⬜ Add negated: true proposals to review YAML for confirmed cases.
  • ⬜ Fix the DnaJ ATP-binding class at the InterPro2GO mapping level; file a GO tracker issue for the phosphorylation class.
  • ⬜ Systematic pseudo-enzyme screen (degenerate active-site motifs).

Read more: projects/TOP_NOTS.md · related projects/PSEUDOENZYMES.md, projects/PAINT.md