Evo2VED: a database of DNA large language model-derived scores for variant interpretation
Tool / method
Database of variant effect scores derived from the Evo2 DNA language model, integrated with biobank GWAS
Summary
Evo2VED (https://dnallmved.org/) is a public database of precomputed variant effect scores from the Evo2 DNA language model, defined as allelic score differences, for 6.48 million human variants common in the gnomAD non-Finnish European population. It provides scores from the 7B and 40B models under three scoring schemes, with annotations of variant function, evolutionary constraint, allele age and disease associations, and integrates GWAS data from UK Biobank, FinnGen and the Million Veteran Program. Three modules support variant-level interpretation, regional exploration and trait-level analysis, and a guided workflow links a GWAS signal to associated regions and variant-level evidence. Case studies of hypertension and depression illustrate candidate variant prioritisation. The scores are evaluated against massively parallel reporter assays, molecular QTLs, fine-mapping probabilities and clinical annotations, including their added value beyond conservation.
Synthesis written by Geno'X. For the full original abstract, please refer to the source publication.
Analysis
This is a resource for complex-trait genetics, not a Mendelian diagnostic tool: it covers only common variants from a single population. The abstract announces an evaluation of Evo2's added value beyond conservation without reporting the result. It should be read alongside this week's zero-shot benchmark, in which Evo 2 differentiates ClinVar variants less well than large multi-species NT models.
Analysis by Dr Thibaut Benquey
Why this score?
Clinical impact: 1/3 · Evidence strength: 2/3 · Novelty: 1/2 · Sample size: 0/1 · Publication status: 1/1 → Total: 5/10
Keywords
Every Wednesday · Annotated selection · Free · Unsubscribe anytime