Google DeepMind Launches AlphaGenome 2 for Cell-Level Genetic Variant Analysis
Google DeepMind has presented AlphaGenome 2, an AI model that evaluates genetic variants for individual cell types rather than only for tissues as a whole. The model is available for non-commercial research via an interface.
Google DeepMind has presented AlphaGenome 2, an AI model for genomic research that no longer evaluates genetic variants only at the level of whole tissues, but for individual cell types. With this, the company is shifting the focus of its genomics work from broad predictions to a much finer resolution.
Previous approaches often grouped many different cells together into a shared tissue profile. However, if a variant acts only in a specific cell type, this average can blur its effect. By modeling the effects of genetic variants per cell type, AlphaGenome 2 is intended to make such differences visible.
According to the available information, the system was trained on more than 107 million cell profiles. Google DeepMind reports that the model thereby identifies almost twice as many disease-relevant gene alterations as its predecessor. Independently verified results on this are reportedly not yet available, and the sources also leave open which diseases and which cell types are specifically covered.
Access is initially restricted to non-commercial research and takes place via a programming interface, known as an API. It is also planned to integrate the model into Google's agentic platform Antigravity. Details on when and to what extent this integration will take place are currently not available.
The development fits into the growing role of artificial intelligence in biomedical research. Whether AlphaGenome 2 actually delivers the claimed progress in detecting disease-relevant changes will likely only become clear once experts have independently reviewed and published the results.
Source: the-decoder.com
13
