Evo2VED: a database of DNA large language model-derived scores for variant interpretation.

Evo2VED provides precomputed Evo2-derived allelic score differences for 6.48 million human variants in the gnomAD non-Finnish European population, enabling variant interpretation with model-based scores.

• Why it matters: Interpreting human variants requires scalable evidence beyond conservation and regulatory features, and researchers need accessible, context-rich outputs from DNA large language models.
• What they did: The database compiles Evo2 scores from the 7B and 40B models under three scoring schemes, adds functional and evolutionary annotations, and integrates GWAS data from UK Biobank, FinnGen, and the Million Veteran Program.
• The result: It supports variant-, region-, and trait-level queries plus downloadable results, and evaluations using MPRA, molecular QTLs, fine-mapping, and clinical annotations test how Evo2 scores relate to and add value beyond existing annotations.

AI summary of the abstract · may contain errors · not medical advice · read the paper.