Google DeepMind unveils AlphaGenome 2 to predict the impact of genetic variants — The Verge

Google DeepMind has unveiled AlphaGenome 2, an artificial intelligence model for predicting how genetic variants affect specific cell types and states. As The Verge reports, the development is intended to give researchers a more detailed understanding of the link between changes in DNA and the causes of diseases.

Predictions for different cells

Nearly all cells in the human body have the same DNA, but different cell types activate or deactivate different genes. These patterns also change depending on the state of a cell. Therefore, the same genetic variant can have different effects, for example, in heart muscle, skin, or immune system cells.

AlphaGenome 2 was created specifically to account for such differences. The model is intended to predict the consequences of genetic variants for specific cells and their states, potentially identifying disease-related effects that may be unnoticed in broader genome models.

Data from 100 million cells

More than 100 million individual cell profiles from 68 tissues were used to train AlphaGenome 2. DeepMind said the model finds nearly twice as many variants capable of changing gene expression in cells as the company’s previous state-of-the-art model.

More current news is available on the UA.News Telegram channel Telegram.

According to Google DeepMind researcher Adam Gayoso, the key result is the combination of human genetics data with cell biology. This could help researchers study genetic variants identified in large populations in greater detail and link their effects at the cellular level to diseases.

DeepMind genomics lead Žiga Avsec noted that such granularity is important for drug development, since medicines target specific proteins and cell types. At the same time, the new model requires more computing resources than its predecessor, AlphaGenome, and therefore does not currently replace it.

Google is making AlphaGenome 2 available for non-commercial research through the AlphaGenome API. For commercial researchers, the company has launched limited private access.