level: research
google deepmind introduced alphagenome atlas, a platform with predictions for 9 billion single-nucleotide variants covering every possible single-letter change in the human genome. it is the most comprehensive catalogue of how genetic mutations affect molecular biology and is available for academic research through a free website portal. the atlas precomputes alphagenome model predictions at scale, giving researchers a big-picture view of variants across the entire genome.
the atlas includes a 1-petabyte dataset, over 30 times larger than the alphafold database. it provides thousands of molecular effect predictions per variant across hundreds of human and mouse cell types. a new alphagenome variant impact (avi) score combines alphagenome and alphamissense predictions into one number for ranking variants. testing shows best-in-class performance on variant pathogenicity and rare disease benchmarks, and the score works for both coding and non-coding regions.
collaborators used the atlas to find a dnm1 gene variant linked to epileptic encephalopathy and to uncover 22% more non-coding genetic associations in uk biobank data. the resource also includes over 2,500 dna sequence motifs and feature attributions explaining which molecular processes are disrupted. it is available for non-commercial use now and for commercial use on google cloud soon, with the base model already on github and via api.
why it matters: it gives researchers a free, searchable map to quickly rank and interpret genetic variants, speeding up rare disease diagnosis and trait discovery without lab testing every mutation.