The Pulse
Google DeepMind releases AlphaGenome 2 for cell-specific research
AlphaGenome 2 predicts how DNA variants affect individual cell types and states. It is available for non-commercial research and in a private preview for commercial research.

AlphaGenome 2 can predict how DNA variants affect particular human cell types and states, moving beyond its predecessor’s predictions across broad tissues, Google DeepMind said in a post on November 11. The model is available through the AlphaGenome API for non-commercial research and in a private preview for commercial research.
It is intended for research use and has not been validated or approved for clinical or diagnostic purposes. DeepMind says the model can help researchers identify variants that act in disease-relevant cells and make predictions for cell states it has not previously encountered.
A shift to individual cell types
The earlier AlphaGenome predicted molecular activity across broad tissues. The new model takes a million-letter DNA sequence and a biological condition—such as a tissue, cell type or state—and predicts effects for that condition at a time.
That condition can come from a reference catalog or be generated from a scientist’s own single-cell experiment. DeepMind trained the model on bulk and single-cell data from human and mouse samples, including more than 107 million individual cell profiles.
The model predicts gene expression, histone modifications and chromatin accessibility, as its predecessor did, as well as DNA methylation. It can also use one measurement, such as single-cell gene activity, to predict an unmeasured property such as chromatin accessibility.
Tests of variant predictions
In a test using TenK10K, a single-cell dataset covering 28 primary immune cell types, AlphaGenome 2 identified 30% of causal variants, compared with 16% for the earlier approach, according to DeepMind. It also identified directional gene effects at 43%, compared with 23%. The comparison was with the original AlphaGenome fine-tuned on single-cell data and other specialized tools.
DeepMind reported that the model computationally reproduced a previous finding involving a variant in a regulatory switch associated with inflammatory bowel disease. The model singled out the variant and predicted that it increased activity of the ETS2 gene in gut macrophages while leaving other cell types and genes largely unaffected.
New cell states and research access
Researchers can supply a single-cell RNA-seq snapshot of gene activity from a few dozen cells. The model uses that information to make predictions for previously unseen cell states without retraining, according to DeepMind.
In an example using data from a COVID-19 mRNA vaccine study, the model distinguished vaccine-activated immune T cells from resting cells. The company said the model also found that a variant linked to severe COVID-19 disrupted gene activity more strongly in the activated state.
DeepMind has also precomputed predictions for trait-associated variants across hundreds of human cell and tissue conditions. The company says the scores can help researchers identify variants and cell types likely to be involved in complex traits, including height and type 2 diabetes.
Interactive Colab tutorials for expression conditioning are available for non-commercial research alongside the model. DeepMind also plans to release AlphaGenome 2 Skills in Google Antigravity.
Source
Companies in this story
Related stories
- DeepMind Maps 9 Billion Human DNA Variants With AlphaGenome
- Google DeepMind opens two datasets used by WeatherNext 3
- Google DeepMind Maps Six Forest Types at 10-Meter Resolution
- Google DeepMind’s AlphaProof Nexus Proves 53 Open Math Problems
- Google DeepMind Adds SynthID Watermarks to AI-Designed Proteins