🧬 AI Gene Review

Overview

The AI Gene Review tool helps researchers and curators review existing Gene Ontology (GO) annotations using strict, defined criteria. It provides a structured workflow for validating annotations using AI-driven analysis combined with literature research and bioinformatics evidence.

Key Features

Quick Start

Installation

1. Install uv for dependency management
2. Clone and install dependencies:

git clone https://github.com/cmungall/ai-gene-review.git
cd ai-gene-review
uv sync --group dev

Basic Usage

Fetch gene data:

uv run ai-gene-review fetch-gene human TP53

Validate a gene review file:

uv run ai-gene-review validate genes/human/TP53/TP53-ai-review.yaml

Generate validation reports with TSV output:

just validate-all  # Creates reports/validation-all.tsv

Fetch publications for a gene:

uv run ai-gene-review fetch-gene-pmids genes/human/TP53/TP53-ai-review.yaml

Generate statistics report:

just stats          # Generate HTML report
just stats-open     # Generate and open in browser

Workflow Overview

Anti-Hallucination Validation

How It Works

Every ontology term requires both an id (semantic identifier) and label (human-readable name):

term:
  id: GO:0005515      # Ontology identifier
  label: protein binding  # Canonical label

Validation Process

The TermValidator performs multi-layer validation:

1. Format Validation: Ensures IDs follow proper CURIE patterns
2. Existence Validation: Verifies terms exist in authoritative ontologies
3. Label Matching: Cross-references provided labels against canonical labels
4. Branch Validation: Ensures GO terms are in correct ontological branches
5. Obsolescence Checking: Flags outdated terms

Why This Prevents Hallucination

• ✅ Dual Verification: Both ID and label must be correct and consistent
• ✅ External Truth Source: Validates against authoritative ontologies
• ✅ Real-time Checking: Uses live API calls to catch fabricated terms
• ✅ Semantic Consistency: Ensures terms make sense in context

Supported Ontologies

The validator supports 10+ major ontologies including GO, HP, MONDO, CL, UBERON, CHEBI, PR, SO, PATO, and NCBITaxon.

Misannotation Analysis

Analysis Framework

Located in analysis/misannotation/ with dedicated tools:

Quick Test Commands:

cd analysis/misannotation

# Quick test with single gene (fast)
just test-quick human TP53

# Full analysis with BLAST (slower)
just analyze-sequence human
just analyze-domains human
just analyze-full human

# Generate summary report
just summarize-risk

Risk Assessment

• HIGH Risk: <25% sequence identity with characterized proteins
• MEDIUM Risk: 25-50% sequence identity
• LOW Risk: >50% sequence identity with consistent domains

Output Formats

• JSON reports: Detailed analysis per gene
• TSV summaries: Easy filtering and analysis
• Risk categorization: HIGH/MEDIUM/LOW classifications

Resources & Links

Gene Review Structure

Each gene review follows a structured YAML format containing:

• Gene metadata: UniProt ID, gene symbol, taxon information
• Description: Comprehensive summary of gene function
• References: Literature and bioinformatics sources
• Existing annotations: Review of current GO annotations with actions
• Core functions: Curated essential gene functions

Example Data

The repository includes example gene reviews for:

• Human: BRCA1, CFAP300, RBFOX3, TP53
• Mouse: Various examples
• Worm: lrx-1

Developer Tools

Available commands using just command runner:

just --list           # Show all available commands
just test             # Run tests, type checking, and linting
just format           # Run code formatting checks
just install          # Install project dependencies
just validate-all     # Validate all genes with TSV output

Case Studies

PedH: a lanthanide-dependent alcohol dehydrogenase (Pseudomonas putida KT2440)

All ten GO annotations on PedH are electronic (IEA), as is typical for non-model genomes. The review weighs each one against the primary literature:

PedH is one of more than 900 P. putida reviews; see the P. putida project.

Epe1: a JmjC protein with no detectable demethylase activity (S. pombe)

The domain assignments place Epe1 among JmjC histone demethylases, and the IBA for histone demethylase activity follows them. The direct evidence does not: the UniProt caution note (the canonical iron-binding His replaced by Tyr at 370), the sequence alignment, and no detectable activity for purified Epe1 in vitro. The IBA is removed as a propagation. The experimental H3K9 demethylase rows are left UNDECIDED, because active-site mutants separate Epe1's functions in vivo and a catalytic contribution is not excluded. The case is catalogued in the IBA review as a pseudoenzyme that kept an ancestral catalytic term.

RIC7 (F4JLB7): a name that points to the wrong gene (Arabidopsis)

F4JLB7 carries the name RIC7, a CRIB-motif ROP effector, but its domain architecture (signal peptide, leucine-rich repeats, hydrophobic C terminus) is consistent with a membrane-associated LRR receptor-like protein, whose ligand and partners remain unresolved. The primers used to clone RIC7 in the original study map to the neighbouring CRIB gene At4g28556, so the experimental annotations (IPI, IMP, IDA) are held as UNDECIDED pending a locus correction.

For work spanning many genes, such as obsoletion migrations, pipeline audits and species campaigns, see the projects index.