DeepMind Releases Massive Genome Dataset to Speed Up Rare Disease Research

DeepMind has released a massive dataset of precomputed predictions for every possible single-letter change in the human genome, called AlphaGenome Atlas. This dataset is a petabyte in size and contains predictions for all 9 billion possible changes.

The predictions are based on a combination of two models: AlphaGenome and AlphaMissense. Each variant gets an AlphaGenome Variant Impact score, which combines the results of both models.

The dataset was validated using whole genomes from over 54,000 UK Biobank participants at Exeter, where researchers found 22% more non-coding associations. The Broad Institute also used the score to surface a variant in the DNM1 gene that creates an incorrect splice site, which was confirmed by experimental screens.

The AlphaGenome Atlas is currently free for academic use, but commercial access will soon be available on Google Cloud, where DeepMind has already sold specialist science models. This raises questions about how European law will treat the two tracks: scientific research and development versus commercial use.