On September 8, 2026, Google DeepMind introduced AlphaGenome Atlas, a database that predicts the molecular effects of all 9 billion possible single-letter changes in the human genome. The company calls it the most comprehensive catalogue of how genetic mutations affect molecular biology.

The Atlas was built by pre-computing predictions from DeepMind's AlphaGenome model across every position in the roughly 3-billion-letter human genome. The result is a 1-petabyte dataset — more than 30 times larger than the AlphaFold Database, according to DeepMind — covering not only single-nucleotide variants but also more than 100 million short insertions and deletions observed in human genomes.

AlphaGenome Atlas vs AlphaFold Database scale
Source: Google DeepMind — Atlas is a 1 PB dataset, more than 30× larger than the AlphaFold Database (relative index).

It is available today through a free website portal that requires no coding, through the AlphaGenome API, and as a skill in Google Antigravity. Access is free for academic and non-commercial research.

Confirmed

The core problem the Atlas addresses is scale. Only about 2% of the human genome codes for proteins; the remaining 98% regulates gene activity and houses most trait-associated variants, yet remains poorly understood. Testing 9 billion mutations in a lab is practically impossible, so DeepMind computed the predictions in advance.

Alongside the raw predictions, DeepMind is releasing the AlphaGenome Variant Impact (AVI) score — a single number per variant that combines AlphaGenome's regulatory predictions with AlphaMissense, the company's model for protein-altering variants. Each AVI score comes with feature attributions showing which molecular processes, such as RNA splicing or gene expression, drive it. The Atlas also includes a compendium of more than 2,500 recurrent DNA sequence motifs mapped across the genome, which DeepMind describes as a step toward a searchable dictionary for non-coding DNA.

Two early use cases come from external collaborators. At the Broad Institute, Laura Covill's team used the AVI score to prioritize variants in an unsolved rare disease case; the tool flagged a variant in the DNM1 gene as creating an incorrect splice site, providing supporting evidence that helped solve the case. Separately, Dr. Gareth Hawkes applied the Atlas to data from more than 54,000 UK Biobank participants and, by grouping variants by predicted molecular effect, uncovered 22% more non-coding genetic associations — including 19 regions linked to body mass index among the top 1% of impactful variants.

DeepMind says its testing shows the AVI score delivers best-in-class performance across many variant pathogenicity and rare disease benchmarks. The underlying AlphaGenome model, published in Nature in January 2026, matched or exceeded the strongest external models in 25 of 26 variant-effect evaluations, according to the paper.

Unknown

The Atlas's predictions are exactly that — predictions. The AVI score's benchmark results come from DeepMind and its academic collaborators, reported today in a preprint that has not yet completed peer review. Independent replication across diverse clinical datasets is still ahead.

Martin Kircher, a bioinformatician at the Max Delbrück Centre for Molecular Medicine in Berlin, cautions that the Atlas will not replace experiments or, for diagnosis, the details of individual cases — though he calls it a useful and generous way to scale up access to a strong model. How the predictions perform on non-European genomes, which are underrepresented in the training data drawn from consortia like ENCODE and GTEx, also remains an open question the launch materials do not address.

Our take

The real story here is distribution, not modeling. AlphaGenome has been available via API since mid-2025, and roughly 9,000 researchers have used it — but that required writing code. Pre-computing everything and wrapping it in a no-code portal removes the computational barrier that kept genome-wide variant scoring out of reach for most rare-disease labs. If the AlphaFold Database's trajectory is any guide, the long-term impact will be measured less by the AVI score's benchmarks than by how many biologists who never touched an API start asking questions they previously could not afford to ask.

Sources