Kansas City, Mo. – September 12, 2026 -- Google DeepMind has released AlphaGenome Atlas, a one-petabyte AI resource containing molecular-effect predictions for more than 9 billion possible single-letter DNA changes across the human genome, giving researchers browser-based access to the largest catalogue of its kind without writing code.
Atlas condenses billions of genetic variants into a single searchable score
The human genome's approximately 3 billion DNA letters create over 9 billion possible single-letter substitutions, a scale impossible to test experimentally in a lab. AlphaGenome Atlas assigns each variant thousands of molecular-effect predictions across hundreds of human cell types and tissues, feeding into a new AlphaGenome Variant Impact (AVI) score that combines AlphaGenome, AlphaMissense and evolutionary conservation data to rank variants across both protein-coding and non-coding regions.
Stowers Institute scientists helped map the genome's regulatory 'dictionary'
Stowers Institute for Medical Research Investigator Julia Zeitlinger, working with Google DeepMind Chief Scientist Žiga Avsec, used the Atlas to analyze regulatory DNA motifs and categorize transcription factors by function -- distinguishing those that alter DNA accessibility from those that activate or repress genes. Stowers bioinformatics scientist Melanie Weilert served as lead author on the project, which is documented in a preprint on bioRxiv.
Zeitlinger, who previously co-developed the BPNet AI framework in 2019 and unveiled the PISA visualization method last month, said cell types operate with subtly different regulatory rules, making general patterns difficult to isolate without a resource capable of querying many cell types simultaneously.
Broad Institute and University of Exeter tested the Atlas on real disease and biobank data
Researchers at the Broad Institute applied the AVI score to prioritize a previously overlooked non-coding variant tied to an unsolved rare disease case. At the University of Exeter, scientists ran the Atlas against genomic data from more than 54,000 UK Biobank participants, identifying additional associations between rare noncoding variants and protein levels. Memorial Sloan Kettering Cancer Center and Stanford University also contributed scientific input on potential applications.
Access is free for non-commercial research, not clinical use
Google DeepMind is offering AlphaGenome Atlas for non-commercial use through its website, stating the predictions are intended to guide research prioritization rather than replace laboratory validation. The company has not validated or approved the tool for clinical applications. Pushmeet Kohli, VP Science at Google DeepMind and Chief Scientist at Google Cloud, said the resource is designed to help scientists worldwide understand what happens when individual genome letters change.