alphagenome-single-variant-analysis

Provide the variant in chr:pos:ref>alt format and the relevant tissue or disease context. The skill scores the variant with dna_model.score_variant across recommended differential scorers, filters significant hits by quantile score, then visualizes REF/ALT track differences and writes a report.

You need an ALPHAGENOME_API_KEY, which you can register for at the DeepMind AlphaGenome page. Store it in a .env file in your home directory; scripts load it automatically via dotenv before creating the dna_client.

Yes, it supports SPLICE_SITES, SPLICE_SITE_USAGE, and SPLICE_JUNCTIONS output types. The interpret_splicing.py script quantifies junction-level changes to detect exon skipping or cryptic splice sites, while analyze_ism.py identifies disrupted motifs.

AlphaGenome does not model protein-coding effects like missense or frameshift consequences, miRNA processing, RNA secondary structure, or developmental and stimulus-specific contexts. High quantile scores with low raw scores in low-expression genes are statistical artifacts indicating no real effect.

score_variant does not accept an ontology_terms argument; you must filter the returned AnnData objects manually by inspecting adata.var columns such as ontology_curie. In contrast, predict_variant does accept ontology_terms directly.

Run resolve_ontology_terms.py with your query term to search the local tissue_ontology_mapping.json file. It returns ranked UBERON, CL, or EFO candidates with fuzzy matching, and you select the most specific match for your disease context.