Google DeepMind released the AlphaGenome Atlas, a precomputed map of every possible single-letter change in human DNA. It covers all 9 billion such changes, plus more than 100 million short insertions and deletions seen in human genomes. The predictions come from AlphaGenome, the model DeepMind released last year. It is free for non-commercial use. Previously researchers had to query the model through a programming interface, which required coding. The atlas adds one summary figure per variant, the AlphaGenome Variant Impact score. DeepMind and academic co-authors report it separated disease-causing mutations from harmless ones in a clinical database. A Broad Institute team used it to flag a variant as a possible cause of severe epilepsy. Researchers also mapped thousands of short DNA motifs and inferred their roles in different cell types. Outside scientists say it will not replace lab experiments or case-by-case clinical judgement.
What changed
Running AlphaGenome across a whole genome required coding against an API and heavy computing power.
What it unlocks
Looking up the predicted effect of any single DNA letter change without writing code.
- 9 billion single-letter changes
- 1 petabyte of predictions
- 100M+ insertions and deletions
- ~9,000 researchers used the API
What you need to act on it
- non-commercial use
Sources