AlphaGenome Atlas opens variant predictions to researchers

Google DeepMind says AlphaGenome Atlas packages predictions for roughly 9 billion possible single-nucleotide variants, covering every single-letter DNA change in the human genome. The company says researchers can browse it through a website portal, use it through the AlphaGenome API, and access it as a skill in Google Antigravity. It also introduced an AlphaGenome Variant Impact, or AVI, score intended to rank variants by likely molecular effect.

The announcement matters because most genetic variants cannot be tested one by one in the lab. DeepMind says the atlas links each variant to thousands of predicted molecular effects, feature attributions explaining what drove the score, and a library of more than 2,500 recurring DNA motifs. It says the dataset is about 1 petabyte, making it larger than a casual download for the lunch break.

What the company says it can already do

DeepMind says external academic collaborators have used the resource in rare-disease work and population genetics. In one collaboration with researchers in the GREGoR Consortium, the AVI score was used to prioritise variants in unsolved rare disease research; the post says experimental screens then validated a DNM1-related splicing effect. In another example, Gareth Hawkes at the University of Exeter applied the atlas to UK Biobank whole-genome data and, according to DeepMind, found more non-coding associations by grouping rare variants by predicted molecular effects.

The company says the tool is available for non-commercial use through the website now, with commercial access on Google Cloud coming later. It also states clearly that AlphaGenome Atlas is not a substitute for medical advice and has not been validated or approved for clinical use.