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.
Continue reading

Latest in AI & Compute

Density limits are no longer one number. They are an engineering moat.

EAST shows density limits can be lifted with start-up control, while W7-X proves long pulses are possible with continuous fueling. Investors should track the enabling stack that makes both possible.

1.3-1.65x

Reported density range above the Greenwald limit (preprint).

0.8-1.0x

Typical EAST density range before the new method.

13 keV

Temperature scale for optimal D-T fusion (about 150M K).