Google DeepMind has released AlphaGenome 2, a new AI model designed to predict how genetic mutations affect individual cell types and biological states. The model serves as a successor to the original AlphaGenome, offering a more granular view of how the same genetic variant can produce different effects depending on whether it is in a heart muscle cell, a skin cell, or an immune cell.
To achieve this level of detail, researchers trained the model on more than 100 million individual cell profiles spanning 68 different tissues. According to Google, the model identifies nearly twice as many variants that change gene expression compared to its predecessor. The company stated that the tool effectively turns static cell atlases into a predictive engine for human genetics.
Researchers found that the model can predict outcomes for new cell types and biological states without requiring retraining. It is also capable of identifying gaps in existing experimental data, which may assist scientists in determining which areas require further laboratory investigation.
“When you develop drugs, you’re targeting specific proteins and specific cell types,” genomics lead Žiga Avsec told The Verge, noting that this level of granularity is vital for disease treatment. Adam Gayoso, a research scientist at Google DeepMind, said the breakthrough lies in "tying human genetics to cell biology" by allowing researchers to see the cellular-level effects of genetic variants identified in large populations.
While AlphaGenome 2 provides more detailed predictions, Avsec noted the model is more expensive to produce computationally than the original version. Consequently, it will not replace the previous model immediately and was not used for the genome-wide Atlas released by DeepMind in September.
Google is making AlphaGenome 2 available for non-commercial research through the AlphaGenome API starting Wednesday. A limited private preview is also being offered for commercial research purposes.
