Google DeepMind launches AlphaGenome Atlas, a genome-wide map of variant effects

Google DeepMind has released AlphaGenome Atlas, a public database that pre-computes the predicted effect of every possible single-letter genetic change in the human genome, giving researchers a free, code-free way to look up variant effects at scale.

What's new

DeepMind describes the tool plainly: "AlphaGenome Atlas is a database that predicts the effects of every possible single nucleotide variant in the human genome." The atlas is built on top of AlphaGenome, DeepMind's sequence model for predicting how DNA changes affect gene regulation, and extends it from a research tool into a browsable public resource.

Key specifics from the release:

• Covers all roughly 9 billion possible single-nucleotide variants across the genome's 3 billion base pairs
• Predictions were pre-calculated into a dataset DeepMind puts at roughly 1 petabyte
• Introduces an AlphaGenome Variant Impact (AVI) score meant to help researchers rank and prioritize which variants are worth follow-up study
• Covers both coding regions (the ~2% of the genome that directly codes for proteins) and non-coding regions, where regulatory variants are far harder to interpret
• Available now at alphagenome.google/atlas as a searchable web portal, with no coding or bioinformatics pipeline required

Context

Most of the human genome is non-coding DNA, and most disease-associated variants found by genome-wide association studies fall in these non-coding regions, where it's historically been difficult to say what a given mutation actually does. AlphaGenome, the underlying model, was built to predict downstream regulatory effects — things like gene expression, splicing, and chromatin state — from raw DNA sequence. Turning that model loose on every possible variant, rather than requiring a researcher to run it on demand, is what makes the Atlas new: it converts a research model into pre-computed, queryable infrastructure.

This lands in the same week DeepMind also published research on 16 AI-for-climate grants across Asia-Pacific, part of a broader push by the lab to point its models at biology, climate, and public-health problems alongside its work on Gemini.

Why it matters

For genomics and clinical research, the bottleneck has increasingly shifted from generating genetic data to interpreting it — knowing which of the millions of variants found in a patient's genome actually matter. DeepMind says the Atlas is intended as exactly that kind of interpretation layer: "AlphaGenome Atlas is already acting as a powerful augmentation partner for the scientific community, accelerating research." A free, pre-computed, searchable index covering effectively the entire space of possible single-letter mutations lowers the barrier for smaller labs and clinical researchers who don't have the compute or expertise to run large sequence models themselves, and could speed up the process of linking specific variants to disease risk.