Deepmind's AlphaGenome Atlas maps every possible DNA change in the human genome
decoder ·
Google DeepMind has released the AlphaGenome Atlas, a comprehensive dataset that predicts the biological effects of nearly all nine billion possible single-letter DNA changes in the human genome. Built on the AlphaGenome AI model introduced in 2025, the atlas offers about 27,000 predictions per variant across hundreds of cell types and tissues, spanning one petabyte—over 30 times the size of the AlphaFold database. To make this practical, DeepMind developed the AlphaGenome Variant Impact Score (AVI), a neural network that condenses the predictions into a single number per variant and outperformed the established CADD tool in tests, especially for noncoding regions. The atlas proved useful in a real case, helping diagnose a child's severe epilepsy by identifying a harmful splice-site variant in the DNM1 gene that only affects brain-specific transcripts. It also enabled researchers analyzing UK Biobank data to find 22 percent more links between noncoding variants and blood protein levels. The tool is intended for research, not direct diagnosis, and is available for noncommercial use through a web portal, API, and Google Antigravity, with a commercial version planned via Google Cloud.
decoder