Skip to content

ai4curation/ai-gene-review / .claude/skills/gocam-curation

gocam-curation

Verified

Evaluate and review GO-CAM (Gene Ontology Causal Activity Model) activities / annotons — molecular-function typing, has-input, causal relations, complexes, evidence — and check their consistency with gene annotation reviews. Use when reviewing cached GO-CAMs under gocams/, filling a GoCamReview YAML (gocams/<id>/<id>-review.yaml), or grounding a module in GO-CAM models.

references/qc-and-consistency.md

View on GitHub
# Verdict, QC flags, consistency & checklist

These controlled vocabularies are the schema enums used in a
`GoCamActivityReview`. Fill them per activity.

## Verdict — `GoCamClaimVerdictEnum`

Overall best-practice/evidence verdict for the activity (the forensic
claim-validation scale):

- **OK** — well supported and best-practice-compliant (correct MF specificity,
  correct `has input`/causal usage, adequate evidence).
- **UNCERTAIN** — defensible but imprecise or under-supported; needs a caveat.
- **WRONG** — contradicts the evidence or breaks a hard rule (binding-as-function,
  reversed causality, etc.).

No "probably fine" — every activity gets an explicit verdict, ideally with a
verbatim quote-back in `supporting_text`.

## QC flags — `GoCamQcFlagEnum`

Record every specific issue observed (multivalued):

| Flag | Meaning | Detail |
|---|---|---|
| `BINDING_AS_FUNCTION` | bare binding MF, no consequence | molecular-function.md |
| `GENERIC_MF` | generic term where a child applies | molecular-function.md |
| `WRONG_MF_TERM` | MF is wrong, not vague (fix may be a parent or a lateral term) | molecular-function.md |
| `HAS_INPUT_MISUSE` | wrong `has input` filler | has-input.md |
| `DIRECT_VS_INDIRECT_CAUSAL` | directness wrong for the mechanism | causal-relations.md |
| `INCORRECT_DIRECTIONALITY` | reversed subject→object edge | causal-relations.md |
| `WRONG_CAUSAL_SIGN` | positive edge where the evidence shows inhibition, or vice versa | causal-relations.md |
| `MISSING_LOCATION` | no `occurs_in` | — |
| `MISSING_PROCESS` | not tied to a process via `part_of` | — |
| `ORPHAN_ACTIVITY` | no causal edges | causal-relations.md |
| `MISSING_EVIDENCE` | no ECO code / reference | — |
| `COMPLEX_SUBUNIT_REPRESENTATION` | wrong subunit/complex attribution | complexes.md |
| `UNRESOLVED_COMPARTMENT` | `occurs_in` reaches no distinct compartment (e.g. chromatin) | compartments.md |
| `NONADJACENT_COMPARTMENTS` | causal edge skips a compartment (e.g. cytosol → extracellular) | compartments.md |

Use no flag rather than a near miss: `verdict` and `source_defects` already
carry the signal, and an inaccurate flag poisons flag queries.

## Consistency with the gene review — `GoCamConsistencyEnum`

How the activity relates to the gene's `genes/**/<gene>-ai-review.yaml` (join on
`gene_product` via `gocams/index.tsv`):

- `CONSISTENT` — matches an accepted core function in the review.
- `MORE_SPECIFIC` / `MORE_GENERAL` — a clean subsumption either way.
- `RELATED` — same area, neither match nor clean subsumption.
- `CONFLICT` — the review **removed / negated / over-annotated** this function.
  Consistency judges the activity's *function*; a wrong causal edge on a
  correct function is `CONSISTENT` plus `WRONG_CAUSAL_SIGN`, not `CONFLICT`.
- `NOT_IN_REVIEW` — function not represented in the review (a candidate gap).
- `NO_GENE_REVIEW` — no review exists yet for this gene product.

`CONFLICT` and `NOT_IN_REVIEW` are the signals worth surfacing.

## Start from the linter

Run `just gocam-lint <model_id>` before reviewing. It lists structural
anomalies (compartment crossings on direct edges, same-gene chains, duplicate
nodes, opposite-sign fan-out, transcription factors with direct edges, ligands
as receptor inputs, generic or binding MFs, orphans, parallel components) with
the QC flag each maps to. Every finding still needs checking against the
literature; see `docs/gocam_lint.md`.

## Pre-finalize checklist

- [ ] MF is the most specific correct term (not generic, not bare binding).
- [ ] `has input` names the substrate/target-gene/effector/cargo — not a ligand or raw DNA.
- [ ] Causal edges are direct/indirect as the mechanism warrants, pointing subject→object.
- [ ] `occurs_in` and `part_of` present where known.
- [ ] Each `just gocam-compartments` lead checked against the evidence, and either flagged with the missing step named or dismissed in `notes`.
- [ ] Every activity/edge has an ECO code + reference; claims have verbatim support.
- [ ] Complex activities attributed to the right subunit(s).
- [ ] `verdict`, any `qc_flags`, and `consistency` recorded for every activity.
receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementationStars: 323kUpdated Claude CodeCodexVerifiedInstall
requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirementsStars: 323kUpdated Claude CodeCodexVerifiedInstall
using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallbackStars: 323kUpdated Claude CodeCodexVerifiedInstall
verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions alwaysStars: 323kUpdated Claude CodeCodexVerifiedInstall