DeepMind Launches AlphaGenome Atlas, Mapping 9 Billion DNA Variants

Google DeepMind introduced AlphaGenome Atlas on September 8, 2026, as a database predicting the molecular impact of every possible single-letter DNA change in the human genome. [1][2]

To create the 1-petabyte dataset, DeepMind used its AlphaGenome AI model and in silico saturation mutagenesis to precalculate the regulatory impact of all nine billion single nucleotide variants. [1][2] The database is free and searchable via a zero-code web portal. [2] To simplify navigation, the Atlas uses a single-number AlphaGenome Variant Impact (AVI) score to combine predictions across both coding and non-coding regions. [1][2]

The underlying AlphaGenome base model, which was released in June 2025 and takes up to one million DNA letters as input, previously demonstrated an ability to show how changes in non-coding DNA disrupt processes like protein production. [1][2] Researchers are already using the precomputed Atlas data to accelerate genetic research. [1]

At the Broad Institute, a team used the AVI score to highlight a critical variant in the DNM1 gene, providing evidence to solve a rare disease case involving an incorrect splice site. [1] Additionally, a researcher using data from over 54,000 UK Biobank participants applied the Atlas to uncover 22 percent more non-coding genetic associations, including 19 regions linked to body mass index. [1]

By shifting the AlphaGenome model from generating individual predictions per API call to providing a fully precomputed 1-petabyte database, DeepMind is drastically lowering the technical barrier for genomic research. This approach allows scientists to query complex molecular impacts using a simple web portal rather than running heavy machine learning infrastructure themselves. The introduction of the AVI score provides a crucial prioritization mechanism for the 98 percent of the genome that remains poorly understood. What the sources don't address: How often the model's in silico variant predictions generate false positives compared to traditional wet-lab experimental validation.