Particle.news
Download on the App Store

DeepMind Publishes AlphaGenome, an AI That Predicts Gene Regulation Across 1 Million DNA Letters

External validation points to research value, not clinical readiness.

Overview

  • DeepMind’s Nature paper formally introduces AlphaGenome and releases the latest noncommercial research version and source on GitHub alongside API access.
  • AlphaGenome analyzes sequences up to 1 megabase and predicts 5,930 human and 1,128 mouse signals, matching or outperforming specialist models in 25 of 26 evaluations.
  • Since a June 2025 preprint, roughly 3,000 researchers across 160 countries have used the tool, with usage at about one million API calls per day, according to DeepMind.
  • The Wellcome Sanger Institute tested the model on roughly 500,000 new experiments and reported strong results, while cautioning that limited and uneven biological datasets cap performance.
  • Researchers deploying the tool at an Undiagnosed Hackathon used it to prioritize noncoding variants in rare disease cases, highlighting its current role in guiding experiments rather than diagnosing patients.