Google DeepMind has launched AlphaGenome Atlas, an AI-powered tool designed to map how every possible single-letter change in the human genome could affect biology, with the aim of speeding up disease research and the development of new treatments. The tool matters because working out which of the countless possible DNA mutations are harmless and which contribute to disease has long been a major bottleneck in genetics and medicine, and DeepMind says its new catalogue is the most comprehensive resource yet for tackling that problem.

The human genome contains around three billion pairs of DNA "letters", and Atlas provides predictions for roughly nine billion possible single-letter substitutions, covering how each might alter molecular processes such as protein production. It builds on DeepMind's earlier AlphaGenome and AlphaMissense models, extends predictions to non-protein-coding regions of DNA, and introduces a "Variant Impact Score" to help researchers prioritise the most significant mutations. The resulting dataset is about one petabyte in size, and Google is making it freely available to researchers for non-commercial use immediately, with commercial access via Google Cloud to follow; the launch comes as DeepMind co-founder Demis Hassabis increasingly focuses on scientific applications, including work at drug-discovery spin-off Isomorphic Labs.