Particle.news
Download on the App Store

DeepMind Publishes AlphaGenome to Predict DNA Function Over Million-Base Spans

Peer review in Nature underscores performance gains, cementing noncommercial availability.

Overview

  • AlphaGenome analyzes up to 1 million DNA base pairs with single-base resolution and predicts 5,930 human or 1,128 mouse functional genomic signals.
  • In benchmark evaluations reported in Nature, the model matched or outperformed prior tools in 25 of 26 tests of variant effect prediction.
  • DeepMind has opened noncommercial access via API since June 2025 and released code on GitHub, with roughly 3,000 researchers in 160 countries making about one million API calls.
  • Researchers describe potential uses in prioritizing cancer-driving mutations, studying rare diseases, discovering drug targets, and designing sequences for gene therapy.
  • Experts caution it is a research tool rather than a clinical diagnostic, noting difficulties with individual-level predictions, human–mouse training limits, and data quality and standardization gaps.