AlphaGenome Atlas turns variant prediction into research infrastructure
DeepMind's precomputed predictions for nine billion possible single-letter variants, now integrated into Ensembl VEP, lower the cost of genome-wide triage. The investor gate is whether a convenient score produces independent, reproducible experimental lift without obscuring uncertainty or provenance.
What to watch
- AlphaGenome Atlas precomputes molecular-effect predictions for nine billion possible single-nucleotide variants in a one-petabyte dataset.
- EMBL-EBI integrated the AlphaGenome Variant Impact score into Ensembl VEP one day after the Atlas launch.
- The decisive diligence metric is prospective experimental lift, not database size or retrospective case studies.