Google DeepMind on Wednesday announced AlphaGenome 2, an artificial-intelligence model that predicts how DNA sequences and genetic variants affect specific types of human and mouse cells, a step the company said could help researchers identify cellular drivers of disease.
Nearly every cell in the human body carries the same DNA, but different cell types switch different genes on and off, and the same genetic variant can have different effects depending on the cell, The Verge reported. Non-coding DNA sequences, which do not code for proteins, make up about 98% of the human genome and are among the hardest to interpret, Nature reported.
DeepMind said the model was trained on measurements from more than 100 million individual cell profiles across 68 tissues, including more than 107 million profiles from human and mouse samples, according to the company’s blog. Nature reported it was trained on 637 human and 311 mouse cell types, and that researchers can supply new single-cell data, including from disease states, to generate predictions without retraining.
The company said AlphaGenome 2 identifies nearly twice as many variants that change gene expression in cells as its previous model, AlphaGenome, released last year, which made predictions across broader tissue averages rather than individual cell types. On a large single-cell map of 28 primary immune cell types known as the TenK10K dataset, DeepMind said the new model identified nearly twice as many causal variants (30% versus 16%) and directional gene effects (43% versus 23%) as the earlier model fine-tuned on single-cell data.
The researchers said the model computationally reproduced aspects of a 2024 finding on inflammatory bowel disease, in which a DNA variant in a distant regulatory sequence overactivates the gene ETS2 in gut immune cells. James Lee, an immunologist and geneticist at the Francis Crick Institute in London who led that study, told Nature: “They’ve managed to figure out the answer in 30 minutes, and it took us a year and a half of experiments.” He added, “It’s reassuring from our perspective.”
Žiga Avsec, who leads DeepMind’s AlphaGenome team, told Nature: “If you want to understand variants associated with genetic diseases, then you have to ask yourself what are the cell types that are affected.” He told The Verge that such granularity matters for drug development because “when you develop drugs, you’re targeting specific proteins and specific cell types.” Avsec also said the more detailed predictions are “more expensive” to produce and that the new model is not replacing the earlier one soon.
DeepMind said the model is available for non-commercial research through the AlphaGenome API, with a private preview for commercial research. The company said AlphaGenome and AlphaGenome 2 are intended for research use only and have not been validated or approved for clinical or diagnostic purposes.