Google DeepMind released AlphaGenome Atlas on September 8, a searchable resource containing predictions about the molecular effects of roughly nine billion possible single-letter changes in human DNA. The Atlas precomputes results from AlphaGenome, a model first introduced in 2025, so researchers can explore changes across the genome without running a separate prediction for every variant.
A new AlphaGenome Variant Impact score, or AVI, ranks changes by their predicted influence on biological processes. Researchers can inspect supporting signals such as gene expression and RNA splicing, including changes in regions that do not code directly for proteins. DeepMind says the Atlas covers about a petabyte of data and includes predictions across numerous tissues and cell types.
The company describes early research collaborations in which the scores helped narrow down possible variants in rare-disease studies, followed by experimental work on selected candidates.
The Atlas website is available for noncommercial research. Google said commercial cloud access was planned, rather than generally available at the September launch. Its predictions are not validated for clinical use and do not provide a medical diagnosis.