Google DeepMind has launched the AlphaGenome Atlas, a massive dataset containing precomputed molecular effect predictions and AVI scores for 9 billion human DNA variants. The project aims to overcome the limitations of running large-scale models on demand, providing a lookup table that accelerates genome-scale research.
The resulting dataset occupies 1 petabyte of storage, making it over 30 times larger than the AlphaFold Database, which contains 200 million protein structure predictions. The system provides AVI scores that demonstrate high performance across benchmarks for variant pathogenicity and rare diseases.
The Atlas is currently accessible for non-commercial research through a dedicated portal and API. While commercial access via Google Cloud is expected soon, the underlying AlphaGenome model, originally released in June 2025, is already available for academic use on GitHub and for commercial use on Google Cloud Model Garden.
