Google DeepMind unveiled AlphaGenome 2 on 11 Nov 2026. The model predicts how DNA changes influence gene activity in specific cell types and states, something earlier tools could only do at the tissue level. Researchers can now ask, for a given variant, what happens in a heart‑muscle cell versus a gut immune cell. The company is offering the model through the AlphaGenome API for non‑commercial research today, and a private preview for commercial projects.
Why finer‑grained predictions matter
Every cell in the body carries the same DNA, but each cell turns different genes on or off. A variant that is harmless in skin cells might disrupt a gene in brain cells, driving disease. By conditioning on a cell’s identity, AlphaGenome 2 can spot such effects that bulk‑tissue models miss. DeepMind’s genomics lead Žiga Avsec says this granularity is crucial for drug development, where therapies target proteins in particular cell types.
How AlphaGenome 2 improves on its predecessor
AlphaGenome 1 used a parallel architecture that spread its capacity across many tissues at once, limiting sensitivity to subtle cell‑type differences. AlphaGenome 2 redesigns the model as a conditional system: it takes a DNA sequence and a specific cell condition as inputs at every layer. Trained on more than 100 million single‑cell profiles from 68 tissues, the new model identifies nearly twice as many expression‑changing variants as AlphaGenome 1. It can also predict effects for cell states it has never seen, such as vaccine‑activated immune cells, without retraining.
Why it matters
For biomedical scientists, AlphaGenome 2 offers a way to link disease‑associated DNA variants directly to the cell types that drive symptoms, potentially speeding up target discovery. For drug developers, the model can help prioritize which variants to address in a specific tissue, making pre‑clinical work more focused. The commercial preview means biotech firms can start testing the tool, though the higher computational cost means it won’t replace the cheaper AlphaGenome 1 for broad surveys.
