Google’s AlphaGenome 2 Moves AI Closer to Explaining How DNA Causes Disease

Google DeepMind has introduced AlphaGenome 2, an AI model designed to predict how DNA variations affect specific human cell types. It targets one of biology’s hardest questions: why can the same genetic change matter in one kind of cell, such as an immune cell or heart-muscle cell, but not another?

Every cell in the human body carries essentially the same DNA, yet cells behave differently because they activate different genes. Previous models could predict the effects of DNA variants across broad tissues. AlphaGenome 2 is designed to work at a much finer level, predicting effects within specific cell types and cell states.

The model was trained on more than 100 million individual cell profiles across 68 tissues from human and mouse data. According to Google, it identifies nearly twice as many variants that alter gene expression in cells compared with its previous state-of-the-art model. On a dataset covering 28 primary immune-cell types, it reportedly identified causal variants 30% of the time, compared with 16% for the earlier model.

What makes AlphaGenome 2 interesting is its conditional design. Scientists can give it a DNA sequence alongside information about a particular cell type or biological condition. The model can then estimate how that DNA behaves in that context. It can also infer missing experimental data—for example, using a snapshot of gene activity to predict other characteristics such as chromatin accessibility.

This could help researchers prioritize which genetic variants are most worth studying in the lab, especially in complex diseases where many small DNA changes may contribute to risk. DeepMind also demonstrated how the model could computationally reproduce a finding involving inflammatory bowel disease and a variant that affects immune cells in the gut.

The company emphasizes that AlphaGenome 2 is a research tool, not a medical diagnostic system. It has not been validated or approved for clinical use. That distinction matters: a powerful prediction model can guide scientific investigation, but it should not be mistaken for direct medical advice.

Still, this is a major example of AI becoming a research partner rather than merely a text generator. The most exciting impact may be invisible to consumers: faster hypothesis-building, better-targeted experiments, and a clearer path toward understanding diseases rooted in genetics.