An AI-powered drug target validation assistant that scores and ranks therapeutic targets against diseases using real-time data from Open Targets Platform and supplementary biomedical databases.
Given one or more gene targets and diseases, this tool:
- Resolves gene symbols → Ensembl IDs and disease names → EFO IDs
- Fetches evidence from Open Targets Platform (genetic associations, clinical trials, tractability, pathways, safety)
- Scores each target–disease pair on three dimensions (0–5 each):
- Clinical Evidence — GWAS hits, rare variants, clinical trial history
- Druggability — small-molecule/antibody tractability, existing chemical matter
- Pathway/Biology — pathway membership, tissue expression, animal models
- Ranks by weighted composite score with compact narrative
- Generates an interactive HTML scoring matrix plus Markdown report, CSV, and raw JSON
- Claude Code installed (
npm install -g @anthropic-ai/claude-code) - An Anthropic API key or Claude Pro/Max subscription
- Python 3.10+ (for the helper script; Claude Code can also query APIs directly)
- No additional API keys needed — Open Targets Platform is free and open
# Clone or download this project
git clone https://github.com/romainstuder/target-ai target-validation-tool
cd target-validation-tool
# Start Claude Code
claudeThat's it. Claude Code reads the CLAUDE.md file on startup and becomes your target validation assistant.
Once inside Claude Code, use these commands:
validate BRAF,KRAS — melanoma,lung cancer
validate PCSK9,HMGCR,ANGPTL3 — cardiovascular disease
validate CDK4,CDK6 — breast cancer, ovarian cancer
validate EGFR,HER2,HER3,MET — non-small cell lung cancer
compare-targets JAK1,JAK2,JAK3,TYK2 — rheumatoid arthritis
find-targets Alzheimer's disease
find-targets Crohn's disease 20
You can also just talk to Claude naturally:
> What are the best drug targets for type 2 diabetes? Score the top 10.
> Is TREM2 a good target for Alzheimer's? Compare it against BACE1 and BIN1.
> I'm working on a grant proposal for targeting GPR84 in inflammatory bowel disease.
Can you validate this target and identify the key risks?
> Score these oncology targets for pancreatic cancer: KRAS, TP53, CDKN2A, SMAD4
The included Python script can also be run independently:
# Search for a target or disease
python3 open_targets_client.py search-target BRAF
python3 open_targets_client.py search-disease "breast cancer"
# Find top targets for a disease
python3 open_targets_client.py find-targets "Alzheimer disease"
python3 open_targets_client.py find-targets "Alzheimer disease" 20
# Full validation
python3 open_targets_client.py validate "BRAF,KRAS" "cutaneous melanoma,non-small cell lung carcinoma"
# Individual lookups
python3 open_targets_client.py target-info ENSG00000157764
python3 open_targets_client.py known-drugs ENSG00000157764
python3 open_targets_client.py association ENSG00000157764 EFO_0000389
# Pathway visualization
python3 open_targets_client.py pathways "PCSK9,HMGCR,ANGPTL3"| Dimension | Weight | Data Sources |
|---|---|---|
| Clinical Evidence | 40% | GWAS Catalog, ClinVar, ClinGen, UniProt, clinical trials (ChEMBL) |
| Druggability | 35% | Open Targets tractability, ChEMBL drugs, structural data |
| Pathway/Biology | 25% | Reactome, literature mining (EuropePMC), expression (Expression Atlas) |
Composite = (Clinical × 0.40) + (Druggability × 0.35) + (Pathway × 0.25)
| Level | Criteria |
|---|---|
| High | Composite ≥ 3.5 with data from ≥ 3 independent source types |
| Medium | Composite 2.0–3.49 or limited independent sources |
| Low | Composite < 2.0 or based primarily on literature mining |
After each validation run, results are saved to the results/ directory. File names encode both targets and diseases:
| File | Contents |
|---|---|
validation_{targets}_vs_{diseases}.html |
★ Interactive HTML scoring matrix — open in browser |
validation_{targets}_vs_{diseases}.md |
Markdown report with tables and narrative |
scores_{targets}_vs_{diseases}.csv |
Machine-readable scores for downstream analysis |
raw_data_{targets}_vs_{diseases}.json |
Raw API responses for reproducibility and auditing |
Example: validation_braf_kras_vs_cutaneous_melanoma_non-small_cell_lung_carcinoma.html
target-validation-tool/
├── CLAUDE.md # Core instructions (read by Claude Code on startup)
├── README.md # This file
├── open_targets_client.py # Python API client, scoring engine + HTML generator
├── report_template.html # HTML template for interactive scoring matrix
└── results/ # Generated outputs (gitignored)
├── validation_*_vs_*.html # ★ Interactive HTML (primary deliverable)
├── validation_*_vs_*.md
├── scores_*_vs_*.csv
└── raw_data_*_vs_*.json
Note: All commands (
validate,find-targets,score-target,compare-targets) are handled viaCLAUDE.mdinstructions — just type naturally and Claude will run the right Python command.
Edit the composite formula in CLAUDE.md under "Step 4 — Compute Composite Score":
# Default: Clinical-heavy (industry standard)
Composite = (Clinical × 0.40) + (Druggability × 0.35) + (Pathway × 0.25)
# Academic / early discovery: Biology-heavy
Composite = (Clinical × 0.25) + (Druggability × 0.25) + (Pathway × 0.50)
# Repurposing focus: Druggability-heavy
Composite = (Clinical × 0.30) + (Druggability × 0.50) + (Pathway × 0.20)
Add new sub-scores to the rubric tables in CLAUDE.md. Claude Code will automatically incorporate them.
Describe additional APIs or databases in CLAUDE.md under "Data Sources" and Claude Code will query them alongside Open Targets.
- Scores are heuristic, not predictive. They summarize available evidence but don't guarantee clinical success.
- Data freshness. Open Targets releases quarterly. Very recent trial results may require supplementary web search.
- Bias toward well-studied targets. Under-researched targets will score lower due to data scarcity, not necessarily due to poor biology.
- Not a substitute for expert review. Use this tool to prioritize and generate hypotheses, not as a final decision-maker.
MIT — use freely for academic and commercial research.
- Open Targets Platform for the comprehensive, open-access target–disease association data
- Anthropic Claude Code for the agentic coding framework
