Google DeepMind has unveiled AlphaGenome 2, a new AI model that predicts how changes in DNA affect gene activity in individual cell types — a finer-grained view of the genome that the company says finds nearly twice as many likely disease-linked variants as its predecessor.

Announced on November 11, the model builds on AlphaGenome, DeepMind's first DNA-sequence model, released in 2025. That earlier system predicted how genetic variants change molecular activity across broad tissues, averaging signals across many different cell types at once. AlphaGenome 2 instead conditions every prediction on a specific tissue, cell type or cell state — the level at which diseases actually take hold.

From tissues to single cells

The new model reads a one-million-letter stretch of DNA together with a single target condition, such as a heart-muscle cell or an immune cell fighting an infection, and makes its prediction for that condition only. The condition can be chosen from a reference catalogue of cell types or generated on the fly from a researcher's own single-cell experiment, using a built-in cell encoder.

That design matters because the same stretch of DNA is read differently in different cells. A variant that is harmless in one cell type can change gene expression in another, and tissue-wide averages can dilute the signal until it disappears. DeepMind trained AlphaGenome 2 on more than 100 million individual cell profiles across 68 tissues, in both humans and mice, to learn those differences.

Nearly twice the hits

On TenK10K, a single-cell map of 28 primary immune cell types, AlphaGenome 2 recovered 30 percent of putatively causal expression-linked variants at 90 percent precision, compared with 16 percent for AlphaGenome 1 fine-tuned on single-cell data. It also got the direction of the effect right 43 percent of the time, against 23 percent for the earlier model.

The numbers are DeepMind's own evaluations, not an independent replication, and the accompanying paper is candid about the limits: even at high precision, the model recovers only a minority of the expression-linked and disease-associated variants that statistical analyses nominate as likely causal. It is a sharper lens, not a finished answer.

Free for researchers — with a caveat

AlphaGenome 2 is available today for non-commercial research through the AlphaGenome API, with interactive Colab tutorials for expression conditioning, and in private preview for commercial research. DeepMind also plans to release AlphaGenome 2 Skills for Google Antigravity, its agentic development platform.

Early adopters are already lining up. Genomics PLC chief executive Peter Donnelly said the company plans to incorporate the model into its Mystra platform, and Sarah Teichmann of the University of Cambridge, who works on the Human Cell Atlas, noted that the model adapts directly to new single-cell data without retraining.

DeepMind stresses that both AlphaGenome versions are intended for research use only and have not been validated or approved for clinical or diagnostic purposes. This is not a tool that will read your personal genome. But for researchers hunting the cellular roots of disease — most disease-linked variants sit in the non-coding regions of DNA — knowing which cell a variant acts in could sharpen drug targets considerably.