Google DeepMind Releases AlphaGenome Atlas With Precomputed Molecular Effect Predictions and AVI Scores for 9 Billion Human DNA Variants

Details

AlphaGenome Atlas converts DeepMind's variant-prediction model into a genome-wide lookup table. AlphaGenome (launched June 2025) could predict how a single DNA change affects gene expression or RNA splicing, but required running per variant. The Atlas precomputes all 9 billion single-nucleotide variants and stores the outputs, removing the compute bottleneck for large-scale genomic studies.\n\nFour resources are bundled. First, molecular effect predictions covering gene regulation across hundreds of human and mouse cell types. Second, the AlphaGenome Variant Impact (AVI) score — a single number combining AlphaGenome's regulatory predictions with AlphaMissense (the protein-impact model), giving unified coverage of both the 2% of the genome that codes for proteins and the 98% that does not. Third, interpretable AVI feature attributions that break down each score into contributions from chromatin accessibility, splicing, conservation, and other categories. Fourth, over 2,500 DNA sequence motifs with genomic locations.\n\nThe 1-petabyte dataset is more than 30 times the size of the AlphaFold Database. Access is free for academic research via a web portal and API. Commercial use on Google Cloud is listed as coming soon. DeepMind reports best-in-class performance on variant pathogenicity and rare disease benchmarks, with details in an accompanying technical report.