Google DeepMind launched AlphaGenome Atlas, a database of predicted effects of DNA changes. It covers every possible single-letter change in the human genome. The predictions were computed in advance using the AlphaGenome model. The Atlas adds a single score, called AlphaGenome Variant Impact, per change. That score merges predictions for protein-coding and non-coding stretches of DNA. Google says it helps researchers rank which changes to investigate first. At the Broad Institute, a team used it to flag a variant in the DNM1 gene. Another researcher applied it to UK Biobank data and found more links to body mass index. The Atlas is available now through a website that needs no programming skills.
What changed
Researchers ran the AlphaGenome model themselves, variant by variant.
What it unlocks
Looking up a predicted effect for any single DNA letter change without writing code.
- 9 billion single-letter changes
- 1-petabyte dataset
- 22% more non-coding associations
- 54,000+ UK Biobank participants
Sources