Google’s AlphaGenome 2 zooms in on the cellular roots of disease
Reported by Nature, The Verge, Google DeepMind and 2 more outlets5 sources ↓
Google DeepMind announced AlphaGenome 2, a model that predicts how DNA variants affect gene regulation in particular cell types and states. It is available for non-commercial research and in private preview for commercial research.
Key points
- AlphaGenome 2 predicts variant effects for one selected cell type or state at a time
- Google DeepMind says it was trained on more than 107 million cell profiles across 68 tissues
- On a 28-cell-type immune benchmark, it identified 30% of candidate causal variants, versus 16% for a fine-tuned predecessor
- The model is available through an API for non-commercial research; commercial research access is in private preview
- Neither AlphaGenome version is validated or approved for clinical or diagnostic use
Sources
- NatureRead ↗DeepMind's ultra-powerful AI for genome mapping wows scientists
- The VergeRead ↗Google’s AlphaGenome 2 zooms in on the cellular roots of disease
- Google DeepMindRead ↗AlphaGenome 2: Deciphering how cells read DNA to better understand disease
- Unite.AIRead ↗Google DeepMind’s AlphaGenome 2 Predicts Cell-State Variant Effects
- The DecoderRead ↗Google Deepmind shifts genomics AI from bulk tissue predictions to single cell type resolution with AlphaGenome 2
Details
AlphaGenome 2 is designed to predict how DNA variants affect gene activity in specific cell types and states, rather than across broad tissue averages. Google DeepMind says researchers can also provide single-cell data from a new experiment to condition predictions on cell states the model has not seen before, without retraining.
The model was trained on more than 107 million individual cell profiles across 68 tissues, according to Google DeepMind and The Decoder. Its predicted readouts include gene expression and chromatin accessibility; the company says it also predicts DNA methylation. It can use one measured property, such as gene activity, to estimate other properties that were not measured.
Unite.AI reports that the comparison was made at 90% precision. Nature reports that the model computationally reproduced aspects of an experimentally established inflammatory bowel disease mechanism involving ETS2 and a regulatory DNA variant.
The model is available for non-commercial research through an API, with commercial research access in private preview. Unite.AI reports that the models are restricted to research and have not been validated or approved for clinical or diagnostic use; it also notes that the model recovers only a minority of variants nominated as likely causal and that its cell-state representations reflect correlations, not causal relationships.
Written from coverage as of
Images
- Google DeepMind
- Nature
- The Verge
- The Decoder
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh