Google DeepMind Introduces AlphaGenome Atlas for DNA Variant Predictions

Google DeepMind published Google DeepMind's AlphaGenome Atlas announcement on September 8, 2026, introducing a platform that contains molecular predictions for 9 billion single-nucleotide variants in the human genome.

The platform precomputes effect predictions for every single-letter mutation. It contains a 1-petabyte dataset, making it more than 30 times larger than the AlphaFold Database expanded in 2022. That previous expansion grew protein 3D structure information from around 190K structures to over 200M structure predictions.

AlphaGenome Atlas links variants directly to functional DNA sequences. It provides thousands of molecular effect predictions per variant across hundreds of human and mouse cell types and tissues. To rank genetic changes, the system includes the AlphaGenome Variant Impact (AVI) score. This score combines predictions from AlphaGenome and AlphaMissense into a single number. It measures impacts across the 2% of the genome that codes for proteins as well as the non-coding 98%. Each score connects to feature attributions and a list of over 2,500 DNA sequence motifs.

External collaborators have applied the dataset to genetic research. Scientists at the Broad Institute used the AVI score to pinpoint a variant affecting the DNM1 gene in rare disease research. At the University of Exeter, a researcher evaluated whole-genome data from over 54,000 UK Biobank participants, identifying 22% more non-coding genetic associations. By isolating the 1% of non-coding variants predicted to be most impactful, the researcher identified 19 genetic regions connected to body mass index.

AlphaGenome Atlas is available for academic research through a website portal, an API, and Google Antigravity. Commercial access on Google Cloud is planned for the future.