DeepMind releases AlphaGenome Atlas: AI map of 9 billion DNA mutations

Google DeepMind has unveiled AlphaGenome Atlas, a comprehensive computational resource that models how all 9 billion possible single-nucleotide variants affect human biology. The platform marks a major advance in genomic interpretation, offering researchers free access to predictions covering genetic variation across the entire genome.

The scale of the challenge#

Understanding which DNA changes matter clinically has long been an intractable problem. The sheer volume of possible mutations—roughly 9 billion single-letter alterations in the human genome—makes laboratory testing of each variant prohibitively expensive and time-consuming. Researchers have therefore struggled to distinguish genuinely damaging variants from harmless ones, especially in the 98% of the genome that does not directly code for proteins.

AlphaGenome Atlas compresses this enormous dataset—at 1 petabyte, more than 30 times larger than DeepMind's AlphaFold Database—into an intuitive interface for biologists with no coding experience, similar to how AlphaFold democratized protein structure prediction three years ago.

Key features and capabilities#

The platform combines several layers of information:

• Molecular predictions: Thousands of effect predictions per variant, spanning gene regulation across hundreds of human and mouse cell types
• AVI score: A unified metric that ranks variant impact in both protein-coding and non-coding regions, with DeepMind noting it achieves best-in-class performance on existing pathogenicity benchmarks
• Feature attribution: Breakdowns showing which biological processes—splicing, gene expression, chromatin accessibility—each variant is predicted to disrupt
• Motif catalog: Over 2,500 recurrent DNA sequences and their genome locations, pinpointing regulatory "words" that control gene activity

The system merges predictions from AlphaGenome (which models broad gene regulatory effects) with AlphaMissense (which predicts protein-level impacts), condensing both into a single interpretable score.

Early research wins#

Academic collaborators have already used the tool to solve concrete problems. Researchers at the Broad Institute, working with the GREGoR Consortium, prioritized rare-disease variants using the AVI score and identified a previously overlooked mutation in the gene that causes epileptic encephalopathy—by recognizing it created an abnormal splice site. Experimental validation confirmed the prediction.