Google DeepMind unveils AlphaGenome 2 to predict genetic variant effects on cell types
Successor model trained on 100M+ cell profiles across 68 tissues aims to link genetic variants to disease mechanisms and potential treatments.
1 source · cross-referenced
- Google DeepMind released AlphaGenome 2, a successor model to its genome-prediction AI, designed to predict how genetic variants affect specific cell types and states.
- The model was trained on over 100 million individual cell profiles across 68 tissues, offering a finer-grained view than its predecessor.
- AlphaGenome 2 identifies nearly twice as many variants that can change gene expression in cells compared to its previous state-of-the-art model.
- The model is available for non-commercial research via the AlphaGenome API, with a limited commercial preview.
Researchers at Google DeepMind announced AlphaGenome 2, a new AI model designed to predict how genetic variants affect different cell types and states. The model builds on its predecessor, AlphaGenome, by providing a more detailed view of how the same genetic variant can have varying effects depending on the cell type and its state.
AlphaGenome 2 was trained on more than 100 million individual cell profiles across 68 tissues, enabling it to capture finer-grained predictions of genetic variant effects on gene expression in specific cells. According to DeepMind, the model transforms static cell atlases into a predictive engine for human genetics.
The company reports that AlphaGenome 2 identifies nearly twice as many variants that can change gene expression in cells compared to its previous state-of-the-art model. It can also make predictions about new types of cells and biological states without retraining and identify gaps in experimental data, which may guide future laboratory investigations.
Adam Gayoso, a research scientist at Google DeepMind involved in the project, emphasized the model's ability to link human genetics to cell biology, allowing researchers to examine the cellular effects of genetic variants identified in large population studies. Žiga Avsec, the genomics lead at DeepMind, noted that such granularity is critical for drug development, as treatments often target specific proteins and cell types.
DeepMind attributes the feasibility of AlphaGenome 2 to advances in technologies that capture high-resolution data on individual cells and their states, enabling the creation of the large datasets required for training. The team is working to expand the model by generating more data, including for underrepresented cell states, and exploring training across multiple species for broader biological insights.
Looking ahead, Avsec suggested that models like AlphaGenome 2 could contribute to efforts to create 'virtual cells,' digital simulations of biological systems. However, he noted that AlphaGenome 2's detailed predictions come with higher computational costs compared to its predecessor, making it less suitable for some large-scale tasks, such as producing genome-wide atlases.
Google DeepMind is making AlphaGenome 2 available for non-commercial research via the AlphaGenome API, with a limited 'private preview' for commercial research. The release follows DeepMind's earlier genome-wide Atlas project and aligns with broader efforts by Google and other AI labs to apply AI to scientific and medical challenges.
- Nov 11, 2026 · Google DeepMind — Blog
Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cell types
Trust84 - Nov 11, 2026 · arXiv cs.AI
Researchers demonstrate auditable human-agent teams for autonomous synchrotron X-ray experiments
Trust79 - Nov 11, 2026 · arXiv cs.AI
Sparse autoencoders enable compact, prefill-trained topic monitors that generalize to decoding
Trust79