DeepMind launches AI-powered Atlas mapping every possible human DNA mutation

DeepMind announced AlphaGenome Atlas, an AI-driven resource that predicts how each of the roughly 9 billion possible single-letter DNA variants could affect molecular biology, such as protein production levels. The atlas extends predictions beyond protein-coding regions to regulatory sequences that control gene activity. Users can query the predictions through a dedicated web portal, the Antigravity agentic platform, or the AlphaGenome interface, and a newly introduced Variant Impact Score ranks variants by their likely biological significance.

The system builds on DeepMind’s prior models AlphaGenome and AlphaMissence, which were trained on publicly available human and mouse genome databases. The resulting dataset, estimated at about 1 petabyte, is being released for non-commercial research today, with commercial licensing to follow on Google Cloud. The launch follows DeepMind co-founder Demis Hassabis’s shift toward scientific research, including work at the drug-discovery spinoff Isomorphic Labs.