Overview
- DeepMind published the AlphaGenome Atlas as a searchable, precomputed catalogue of roughly 9 billion single‑nucleotide variants and packaged the results into about 1 petabyte of data to remove the need for users to run large models themselves.
- The Atlas gives each variant an AlphaGenome Variant Impact (AVI) score, a single metric that combines predictions from coding and noncoding regions so researchers can quickly rank which mutations to follow up experimentally.
- DeepMind made the Atlas available for noncommercial research access through a web portal and APIs on Tuesday, and said commercial access via Google Cloud will be offered soon under licensing.
- Early, limited uses reported by DeepMind include a Broad Institute team using the AVI score to prioritize a DNM1 splice‑disrupting variant in a rare disease case and a UK Biobank analysis that flagged a 22% increase in noncoding associations and 19 new BMI‑linked regions, but both findings need broader independent validation.
- The project builds on DeepMind’s prior genomics models and follows the AlphaFold data‑release playbook, and it could speed genetic discovery while raising questions about cloud dependence, licensing, and the need for laboratory confirmation before clinical use.