Skip to content

AI · 12 November 2026 · 3 min read

DeepMind releases AlphaGenome 2 for non-commercial research

Google DeepMind said on 11 November that AlphaGenome 2 predicts how DNA changes affect specific human cell types. It was trained on over 107 million cell profiles across 68 tissues. On the TenK10K immune set it identified 30 percent of causal variants, against 16 percent for the previous model. Non-commercial use is through an API. Commercial use is a private preview.

DeepMind’s AlphaGenome 2 graphic, pink cubes on a dark field beside the model name.
Google DeepMind’s graphic for the AlphaGenome 2 post. It is a graphic. It is not a cell, and it is not a laboratory.

Google DeepMind said on 11 November that AlphaGenome 2 predicts how a DNA sequence, and changes to it, affect specific human cell types. The post says the model was trained on bulk and single-cell data from human and mouse samples, including over 107 million individual cell profiles, and on more than 100 million profiles across 68 tissues. It says a researcher can supply gene-activity data from a few dozen cells so the model can be conditioned on a cell state it was not trained on, without the post describing a separate retraining step.

On the TenK10K dataset, which DeepMind describes as a map of 28 primary immune cell types, the post says AlphaGenome 2 identified 30 percent of causal variants, against 16 percent for the original AlphaGenome after it was fine-tuned on single-cell data. It says the share of directional gene effects was 43 percent, against 23 percent. DeepMind also points to AlphaGenome Atlas, which it says holds predictions for all 9 billion possible single-letter changes in the human genome. The model is available for non-commercial research through the AlphaGenome API, and in private preview for commercial research.