Google DeepMind's AlphaGenome Atlas predicts effects of all 9 billion human DNA variants

A predictive map of the whole genome

Google DeepMind has released AlphaGenome Atlas, a platform that predicts the molecular consequences of every single-letter change that can occur in human DNA. Announced on September 8, 2026, the catalogue spans roughly 9 billion single-nucleotide variants — the complete set of possible one-letter mutations — and is free for academic research through a web portal.

Precomputed predictions at petabyte scale

According to the Google DeepMind blog, the Atlas was built by running DeepMind's AlphaGenome model, an AI system trained to predict how genetic variants perturb biological processes, across the entire genome and storing the output in advance. The result is a dataset of about 1 petabyte, which DeepMind says is more than 30 times larger than the AlphaFold Database of protein structures.

Each variant in the Atlas carries thousands of molecular effect predictions spanning multiple aspects of gene regulation, computed across hundreds of human and mouse cell types and tissues. The release also includes a collection of more than 2,500 recurrent DNA sequence motifs and their genomic locations.

One score to rank 9 billion variants

To make the resource navigable, DeepMind introduced the AlphaGenome Variant Impact (AVI) score. It condenses the regulatory predictions of AlphaGenome and the protein-impact predictions of AlphaMissense into a single number per variant. The score applies to both the roughly 2% of the genome that codes for proteins and the remaining 98% that orchestrates gene activity, where most variants linked to traits actually reside.

DeepMind says its own testing shows the AVI score achieving top performance across a range of variant pathogenicity and rare disease benchmarks. Every score is paired with feature attributions that decompose it into interpretable contributions, such as predicted disruption to RNA splicing, gene expression, chromatin accessibility or sequence conservation.

Early results from research partners

DeepMind reports that external collaborators have already used the Atlas to advance real cases. Working with the GREGoR Consortium, Laura Covill and Anne O'Donnell-Luria of the Broad Institute applied the AVI score to prioritize candidate variants in an unsolved rare disease. The approach surfaced a variant in the DNM1 gene, which is strongly linked to epileptic encephalopathy, and the underlying AlphaGenome predictions explained how it works: the variant introduces an erroneous splice site that yields an abnormally extended protein. DeepMind says experimental screens confirmed the prediction and turned up nearby variants with similar effects.

In a population-scale application, Gareth Hawkes, a Medical Research Council fellow at the University of Exeter, applied the Atlas to whole-genome data from more than 54,000 UK Biobank participants. Grouping rare variants by their predicted molecular effects made weak signals stand out, yielding 22% more non-coding genetic associations. Focusing on the top 1% of impactful variants pointed to 19 genetic regions associated with body mass index, giving researchers concrete targets for follow-up work.

How to access it

AlphaGenome Atlas is available through a website portal that requires no coding skills, through the AlphaGenome API, and as a skill in Google Antigravity. Access is free for academic research.

Why it matters

Testing 9 billion variants in a laboratory is practically impossible, and the non-coding 98% of the genome remains largely uninterpreted. A precomputed, queryable map of variant effects — paired with a single ranking score — converts an intractable search space into something a clinical researcher can scan in a browser, which is significant for rare disease diagnosis where causal variants hide among thousands of candidates.

The distribution strategy also repeats the AlphaFold Database playbook: run a powerful model at scale, publish the predictions openly, and wrap them in a no-code interface. DeepMind credits that earlier database with accelerating discovery across the life sciences, and it is clearly aiming for a repeat in genomics. One caveat is worth keeping in mind: everything in the Atlas is prediction, not measurement, and the company's performance claims come from its own benchmarks. The AVI score points researchers at likely candidates, but as the DNM1 case shows, experimental validation remains the step that closes the loop.