Google DeepMind has introduced AlphaGenome 2, an updated version of its model designed to interpret the human genome. The tool aims to predict the molecular consequences of DNA sequence variants depending on the cell type involved, a key question for understanding how certain mutations contribute to disease.
Most of the human genome does not directly code for proteins, but instead regulates gene activity through complex mechanisms such as gene expression, RNA splicing, chromatin accessibility, and transcription factor binding. AlphaGenome attempts to model these processes jointly and at scale, accounting for differences across cell types, in order to flag variants that are likely to have a functional effect.
This approach extends the strategy DeepMind pursued with AlphaFold, which predicted the three-dimensional structure of proteins. With AlphaGenome, the company is tackling an additional layer of biological complexity: not just the structure of molecules, but how cells read and interpret genetic information depending on their context.
For researchers working on human genetics, this kind of model could help identify non-coding variants implicated in complex diseases, an area where experimental validation remains slow and expensive. DeepMind says it is making the tool available to the scientific community for research purposes, though it has not publicly detailed the specific improvements over the first version of the model.
The announcement fits into a broader trend of AI applied to the life sciences, as major technology labs continue investing in specialized foundation models for molecular biology, with the hope of accelerating biomedical research.