Google DeepMind announced Tuesday the launch of AlphaGenome Atlas, a predictive map of every possible single-letter change in the human genome. The platform contains molecular effect predictions for roughly 9 billion single-nucleotide variants — every single-letter change possible — in a 1-petabyte dataset that stands as the most comprehensive catalogue of how genetic mutations affect molecular biology.
The human genome consists of about 3 billion base pairs, with only around 2% directly coding for proteins. The remaining 98% is non-coding — regulatory instructions that act like switches, controlling when and how strongly other genes turn on. Working out what a single letter swap does to that regulatory layer has historically meant slow, one-variant-at-a-time lab testing. DeepMind's earlier AlphaGenome model could predict how variants impact biological processes and found widespread research use, but querying it variant-by-variant via API was still tedious. The Atlas removes that friction by precomputing the predictions, so researchers can look up any variant instantly.
To help scientists quickly find the most impactful genetic changes, the team also released the AlphaGenome Variant Impact (AVI) score. It condenses the strengths of AlphaGenome and AlphaMissense — the model for predicting protein-altering variant impact — into a single number that works across both coding and non-coding regions. Researchers can rapidly rank variants, and AVI feature attributions reveal which molecular processes, like RNA splicing or gene expression, drive each score. The Atlas also includes a compendium of over 2,500 recurring DNA sequence motifs — the "words" of the genome — and their locations.
The Atlas is already delivering real-world results. Researchers at the Broad Institute used the AVI score to prioritize variants in unsolved rare-disease research, discovering a DNM1 variant linked to epileptic encephalopathy that earlier work had missed, and experimentally confirming it acts through an incorrect splice site. A University of Exeter team applied it to whole-genome data from over 54,000 UK Biobank participants, uncovering 22% more non-coding genetic associations and pinpointing regulatory variants behind proteins including PLA2G7 (linked to aging) and EGLN1 (a vital cellular oxygen sensor).
AlphaGenome Atlas is available today for academic research through a free-to-use website portal at alphagenome.google/atlas, the AlphaGenome API, and as a Google Antigravity skill, with commercial access on Google Cloud coming soon. At over 30 times the size of the AlphaFold Database — which transformed structural biology by expanding protein structure information from 190K experimental structures to 200M+ predictions — the Atlas looks set to do the same for genomics.