Google DeepMind's AlphaGenome 2 predicts cell-level effects of DNA variants

AlphaGenome 2 builds on DeepMind's earlier genome-predicting model, AlphaGenome, which was released last year. The central problem it addresses is that nearly every cell in the human body carries the same DNA, yet different cell types switch different genes on and off, so the same genetic variant can have very different effects depending on the cell's type and state. The new model is designed to capture those differences, and Google says it effectively turns static cell atlases into a predictive engine for human genetics.

The model became possible only recently, according to DeepMind's Žiga Avsec, thanks to new technologies that can capture individual cells and their states at high resolution and create the vast datasets needed for training. Researchers also found AlphaGenome 2 could make predictions about new types of cells and biological states without retraining, and could identify gaps in experimental data. Avsec said future versions could potentially be trained on data across multiple species, and longer term such models could feed into efforts to create virtual cells.

Despite being positioned as a successor, AlphaGenome 2 is not replacing AlphaGenome any time soon. Avsec said its more detailed predictions are more expensive to produce, which would have made it impractical to use for the genome-wide Atlas covering every possible DNA letter change that DeepMind released in September. The launch sits alongside a broader push by Google and rival AI labs into science and medicine, including Anthropic's experiments with drug design and OpenAI's work on mathematics problems.