Google DeepMind has released AlphaGenome Atlas, a publicly available database designed to help scientists identify which genetic variants are responsible for particular traits or diseases. The tool builds on AlphaGenome, an AI model launched last year that predicts how genetic variants affect biological processes, but addresses a key remaining problem: knowing which variants are worth testing in the first place, rather than relying on slow, brute-force experimentation.

The database contains around a petabyte of data covering predictions for nine billion possible nucleotide variations across the human genome, each assigned an "AlphaGenome Variant Impact" score to help researchers rank which variants are most likely to be significant. In one trial with the GREGoR Consortium, the tool helped identify a variant in the DNM1 gene, linked to a severe brain disorder, and revealed how it caused a faulty splice site that led to an abnormal protein. DeepMind cautions the results remain predictions that should improve as its models develop, and the release follows other science-focused projects such as AlphaFold and weather-forecasting models, even as Google separately ramps up huge AI capital spending and reshapes web search with AI-generated answers.