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

The Evo2 Variant Effect Database (Evo2VED; https://dnallmved.org/) is a publicly accessible resource cataloging precomputed Evo2-derived variant effect scores, defined as allelic score differences, for 6.48 million human variants common in the gnomAD non-Finnish European population. Evo2VED includes scores from the 7B and 40B models under three scoring schemes, alongside annotations of variant function, evolutionary constraint, allele age, and disease associations. It integrates genome-wide association study (GWAS) data from UK Biobank, FinnGen, and the Million Veteran Program. Three complementary modules support variant-level interpretation, regional and region-set exploration, and trait-level analysis through flexible queries, interactive visualizations, and downloadable results. Evo2VED also catalogs associations between Evo2 annotations and per-SNP heritability across thousands of traits from these biobanks. A guided workflow connects GWAS with associated regions and individual variant evidence. Case studies of hypertension and depression illustrate how regional comparisons of Evo2 scores and GWAS signals can guide candidate variant prioritization. Evaluations using massively parallel reporter assays, molecular QTLs, fine-mapping probabilities, and clinical annotations assess associations with Evo2 scores alone and their added value beyond conservation and regulatory annotations. By providing accessible scores, contextual annotations, and comparative analyses, Evo2VED facilitates exploration and application of genomic language model predictions in human genetics research.