Google DeepMind has released AlphaGenome Atlas, a free database containing AI-powered predictions for how every possible single-letter change in human DNA could affect biology. The catalog covers approximately nine billion single-nucleotide variants, plus more than 100 million insertions and deletions already observed in human populations. Researchers can now search a web portal to understand genetic mutations without running complex models themselves, potentially transforming how scientists approach rare disease diagnosis and drug discovery.

What Makes This Different From Previous Genetic Tools?

The human genome contains roughly three billion letters written in an alphabet of four chemical bases, commonly abbreviated as A, C, G, and T. Only about 2% of these letters directly code for proteins; the remaining 98% acts as a control center, determining when and where genes activate. Until now, testing all conceivable single-letter changes in a laboratory was practically impossible. Atlas has precomputed the work for researchers, storing results in a database roughly 30 times larger than the AlphaFold protein database, with a total size of one petabyte.

The tool builds on AlphaGenome, an AI model DeepMind released in June 2025, which learned patterns between DNA changes and biological processes by analyzing public databases of human and mouse genomes. Applying those predictions to billions of possible variants produced the massive dataset now available through Atlas.

How Can Researchers Actually Use This Information?

To help scientists navigate billions of possibilities, DeepMind created a summary metric called the AlphaGenome Variant Impact (AVI) score. This single ranking measures how much impact a mutation could have, combining predictions about gene regulation with protein-altering effects. The AVI score outperforms existing methods like AlphaMissense, which focuses only on protein-coding changes, because it also works effectively with non-coding DNA.

For researchers wanting deeper insight, the system breaks down each AVI score into its component parts, showing whether a mutation's effect comes from splicing, conservation, DNA accessibility, or other processes. The database also includes a catalog of more than 2,500 short DNA sequences called motifs, which are the regulatory components that control gene behavior.

Steps to Access and Interpret AlphaGenome Atlas Predictions

• Academic Access: Researchers can search the web portal or use an application programming interface (API) for non-commercial projects at no cost, starting immediately as of September 8, 2026.
• Score Interpretation: An AVI score of 10 places a variant among the 10% most impactful in the genome, while a score of 30 ranks it among the strongest one in a thousand variants.
• Molecular Breakdown: Review the feature attributions to understand whether predicted effects stem from splicing changes, gene expression shifts, or protein alterations before pursuing laboratory validation.
• Commercial Licensing: Organizations seeking commercial use can access Atlas through Google Cloud with a licensing arrangement, though specific terms have not yet been announced.

What Real-World Results Show About Atlas Accuracy?

Early testers have reported promising results. Researchers at the Broad Institute used AVI scores to identify a variant in the DNM1 gene linked to severe epilepsy. The model predicted the mutation would create a false splice site, stretching the resulting protein by 13 amino acids. Laboratory experiments later confirmed this prediction, and the variant was reclassified as likely pathogenic.

"The human genome is a massive search space. We can use it to shrink the haystack," said Gareth Hawkes, a Medical Research Council fellow at the University of Exeter, describing how Atlas helped narrow candidate variants from 526 to just four in one analysis.

Gareth Hawkes, Medical Research Council Fellow at the University of Exeter

In a retrospective test on previously solved rare genetic disorder cases, the AVI score placed the known causal variant among a patient's top 50 candidates 29.5% of the time, compared to 12.5% for CADD, an existing ranking method. When researchers filtered rare non-coding variants by their predicted molecular effect, they identified 22% more associations than previous methods, including regulatory changes affecting proteins linked to aging and oxygen sensing.

Another researcher applied Atlas to whole-genome data from more than 54,000 participants in the U.K. Biobank, hunting for rare non-coding variants affecting blood protein levels. By focusing on the top 1% of non-coding variants with the most predicted impact, the team identified 19 genomic regions associated with body mass index.

What Are the Limitations Researchers Should Know?

Despite its power, Atlas has important constraints. The tool examines only about one million letters around each variant, meaning distant regulatory switches may be missed. Additionally, many diseases involve multiple genetic variants working together rather than a single letter change. Atlas is also a research tool, not a diagnostic tool, and has received no clinical approval. Any predictions must be validated in the laboratory before clinical use.

"Basically it took us some time to really precompute and also analyze this many variants because the space is so big," explained Ziga Avsec, DeepMind's genomics lead, describing the effort required to create the genome-wide catalog.

Ziga Avsec, Genomics Lead at DeepMind

The release represents the latest effort from Google to use artificial intelligence to tackle core problems in science and medicine. It comes as DeepMind cofounder Demis Hassabis steps back from running the AI lab to focus on scientific research, including leading drug-discovery spinoff Isomorphic Labs. The company's best-known work in this area remains AlphaFold, the protein-structure prediction model that won Hassabis and John Jumper the 2024 Nobel Prize in Chemistry.

For the broader scientific community, Atlas promises to accelerate the understanding of genetic diseases and the hunt for possible cures by making comprehensive genetic predictions freely available to academic researchers worldwide for the first time.