locus-to-gene-mapper-skill

Run the bundled script with a trait query, for example python scripts/map_locus_to_gene.py --trait-query "type 2 diabetes" --print-result. It resolves the trait to an EFO term, pulls GWAS anchor variants, gathers L2G, colocalisation, eQTL, and burden evidence, then writes a JSON mapping and markdown summary.

The pipeline queries EFO/OLS for trait resolution, GWAS Catalog for associations, NCBI RefSNP for GRCh37/GRCh38 coordinates, Open Targets for L2G and colocalisation, GTEx for eQTLs, Genebass for rare-variant burden, plus ClinVar, gnomAD, and Human Protein Atlas context.

Yes. Provide seed_rsids in the input JSON, such as ["rs7903146", "rs13266634"], and the pipeline resolves their coordinates via NCBI RefSNP and gathers evidence per locus. You can also combine seed variants with a trait query.

The overall score is a weighted sum: 0.40 times max L2G, 0.25 times max colocalisation h4, 0.15 times eQTL tissue support, 0.10 times burden evidence, and 0.10 times coding support. Scores map to High, Medium, Low, or VeryLow confidence labels.

Association evidence alone cannot prove causality, and the skill explicitly avoids claiming definitive causal genes. Missing evidence is recorded as a limitation and lowers confidence rather than being imputed, and unresolved GRCh38 coordinates mark the run as degraded.

The script requires Python 3.11 or later and the requests package for API calls. Figure generation is optional and additionally needs matplotlib, seaborn, and pandas; without them the pipeline falls back to Mermaid diagrams in the markdown summary.