The AlphaGenome Atlas is a 1-petabyte database that predicts the molecular effects of virtually every possible single-letter change in the human genome
Google DeepMind has released a database that could meaningfully change how scientists study genetic disease. The AlphaGenome Atlas, launched on September 8, maps the predicted molecular effects of roughly 9 billion possible single-nucleotide changes across the human genome, making one of genomics’ most computationally expensive problems into a free lookup table.
What DeepMind actually built #
The Atlas is built on top of AlphaGenome, an AI model DeepMind developed in 2025 to analyze long DNA sequences and predict multiple molecular outcomes simultaneously. That was already a step beyond most existing tools, which tend to focus narrowly on one type of molecular effect at a time.
The new Atlas takes those predictive capabilities and precomputes them at genome-wide scale. At 1 petabyte, it is roughly 30 times larger than the AlphaFold database, which itself redefined structural biology when it catalogued the predicted 3D shapes of hundreds of millions of proteins.
The database covers single-nucleotide variants, the most common form of genetic variation in humans. To help prioritize them, the Atlas includes what DeepMind is calling the AlphaGenome Variant Impact (AVI) score. The score ranks variants by their predicted molecular significance, giving researchers a systematic starting point instead of requiring them to evaluate billions of candidates manually.
A particularly important focus is the non-coding genome, the roughly 98% of human DNA that doesn’t directly encode proteins. This region is frequently where genome-wide association studies find disease-linked variants, yet it’s historically been harder to interpret. The Atlas is explicitly designed to address that gap.
Early results and who’s already using it #
The Broad Institute has already been working with the Atlas and reports faster variant prioritization and new genetic associations identified through analyses of UK Biobank data. The UK Biobank is a large-scale biomedical database containing health and genetic information from around half a million UK participants.
The Atlas is freely available for non-commercial research through a web portal and API, following the same open-access model DeepMind used with AlphaFold. Commercial licensing is planned through Google Cloud.
Why this matters beyond the lab #
By precomputing 9 billion predictions and making them queryable, DeepMind effectively lowers the entry barrier for smaller research groups, universities in lower-resource settings, and rare disease researchers who may be working with limited compute budgets. A rare disease team studying a single gene no longer needs access to a supercomputer to understand what a specific variant in a regulatory region might do.
Biobanks globally are expanding. Whole-genome sequencing costs have fallen dramatically over the past decade. The volume of variant data researchers need to interpret is growing, and the tools to interpret it have not scaled proportionally. A precomputed, standardized reference that can be queried rather than computed fills that gap in a way that individual lab-by-lab model deployment cannot.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our