Google DeepMind has released AlphaGenome Atlas, a searchable database that predicts the molecular effects of every possible single nucleotide variant in the human genome. The resource pre-calculates the likely regulatory impact of all 9 billion single-letter genetic changes, creating a 1-petabyte dataset that DeepMind describes as the most comprehensive catalogue of how genetic mutations affect molecular biology.
The Atlas is designed to make a demanding part of genetic research quicker to query. Researchers previously needed computationally intensive custom analyses to examine many variants. AlphaGenome Atlas instead brings those predictions together in a database that can be searched through a website portal requiring zero coding skills.
Its central measure is the AlphaGenome Variant Impact, or AVI, score. The score combines AlphaGenome and AlphaMissense predictions across coding and non-coding regions, ranking mutations from low to high impact. It also aims to show how a change may cause harm, including by disrupting gene switches or RNA splicing instructions.
The emphasis on non-coding DNA addresses a longstanding limitation in genomic interpretation. Scientists understand relatively well the 2% of the genome that codes for proteins. Knowledge is more limited for the remaining 98%, despite its role in regulating molecular processes. The AVI score is intended to provide a single route through predictions for both parts of the genome.
The dataset is more than 30 times the size of the AlphaFold Database. DeepMind says access will allow clinical researchers and biologists worldwide to investigate variants without having to run the underlying model themselves. Free access for non-commercial research is available through the Atlas web portal, the AlphaGenome application programming interface on GitHub, and a skill inside Google Antigravity.
Commercial availability on Google Cloud is planned, although DeepMind has not given a timeline. The base AlphaGenome model is already available on Cloud Model Garden.
The Atlas builds on the AlphaGenome model, first described in a June 2025 preprint and later published in Nature in January 2026. The foundational model could process up to 1 million base pairs of DNA at once. It predicted outcomes including gene expression levels, RNA splicing patterns and chromatin interactions at single-base resolution.
DeepMind says its testing found that the AVI score reached best-in-class performance across variant pathogenicity and rare disease benchmarks. Early research cited by the company suggests the tool can help identify signals that standard approaches may miss.
At the Broad Institute, Laura Covill's team applied the AVI score to unsolved rare disease cases. The analysis surfaced a variant in the DNM1 gene that the model predicted would create an incorrect splice site. Separately, a UK Biobank study using the Atlas found 22% more non-coding trait associations than standard analyses.
Those findings offer examples of how pre-computed predictions could help researchers prioritise which genetic changes deserve further investigation. They do not establish the Atlas as a diagnostic system. DeepMind says AlphaGenome has not been validated for, or approved for, clinical use. It describes the service as a research tool for hypothesis generation and prioritisation.
Whether the Atlas will gain clinical validation or regulatory approval remains unknown. Its eventual usefulness across particular diseases and conditions will also depend on further research. For now, the release gives researchers a large, freely accessible set of AI-generated predictions covering the full genome, including the regions where genetic effects have been harder to interpret.
Comments (0)