Google DeepMind Releases 1PB AlphaGenome Atlas for 9 Billion DNA Variants

Google DeepMind released AlphaGenome Atlas, a one-petabyte dataset that predicts the molecular effects of 9 billion possible single-letter DNA changes across the human genome. [1][3] The resource is available immediately for non-commercial research through a free web portal, while commercial access will follow soon through licensing on Google Cloud. [3][4] According to DeepMind vice president for research Pushmeet Kohli, the release allows researchers to access a comprehensive map of human genetic variation using a web browser. [4] To build the Atlas, DeepMind researchers ran the AlphaGenome AI model, which was released in June 2025, across a reference version of the human genome. [4][5] The resulting dataset is more than 30 times larger than the AlphaFold Database, a previous release that holds over 200 million protein structure predictions. [5]

The AlphaGenome Atlas provides the AlphaGenome Variant Impact (AVI) score, a single number that ranks variants by their predicted biological impact. [3][5] This metric condenses predictions by combining regulatory data from AlphaGenome with AlphaMissense, DeepMind's model for assessing protein-altering DNA variants. [3][5] Because it integrates these models, the AVI score functions across both protein-coding DNA—which comprises about two percent of the genome—and the much larger non-coding regulatory regions. [4][5] Instead of researchers running predictions individually, they can use the database to access roughly 27,000 separate predictions attached to each individual variant. [4] The tool is also available through the AlphaGenome API and as a skill in Google Antigravity. [5] A paper detailing the development of the Atlas was posted as a preprint on bioRxiv. [4]

The AlphaGenome Atlas shifts the computational bottleneck in genetics by offering precomputed outputs instead of requiring researchers to run large models on demand for each candidate variant. While the underlying AlphaGenome model is already available on GitHub for academics and on Model Garden for commercial use, the precomputed Atlas vastly expands the model's reach. The licensing structure indicates Google's strategy to monetize the data, as even its sister company Isomorphic Labs will need a commercial license to access the Atlas through Google Cloud. This setup contrasts with the immediate free access granted to academic researchers through the web portal. What the sources don't address: Whether the initial commercial release on Google Cloud will include specific tools tailored for enterprise pharmaceutical workflows.