Google DeepMind’s AlphaGenome 2 Predicts How DNA Mutations Affect Specific Cells

Google DeepMind has introduced AlphaGenome 2, a next-generation deep-learning model designed to predict how DNA sequences and their variations impact specific human cell types. This new iteration gives researchers a powerful tool for uncovering the cellular root causes of complex diseases.

While the first version of AlphaGenome evaluated molecular activity across wide tissue samples, it often missed critical signals hidden within specific, specialized cell types. In biology, every cell in the human body contains the same core DNA blueprint, yet a heart muscle cell reads those instructions completely differently than an immune cell.

AlphaGenome 2 addresses this challenge by functioning as a conditional model. Instead of producing a single fixed output for an entire tissue chunk, it evaluates one biological condition, cell type, or cell state at a time. Trained on massive datasets covering more than 107 million individual cell profiles across 68 tissues from human and mouse samples, the model turns static cell atlases into a dynamic predictive engine.

The model takes a stretch of DNA roughly one million letters long alongside a target cellular condition. This design enables scientists to input single-cell gene activity snapshots from new laboratory experiments, allowing the system to adapt to fresh data without requiring full retraining.

Furthermore, AlphaGenome 2 supports cross-modality imputation. If researchers provide basic gene-expression measurements, the model can predict unmeasured properties, such as chromatin accessibility or DNA methylation, helping teams prioritize which physical experiments to run next.

According to DeepMind’s internal benchmarks, the updated system identifies nearly twice as many expression-altering genetic variants in specific cells compared to previous state-of-the-art tools.

Google DeepMind has made AlphaGenome 2 available starting today for non-commercial research use via the AlphaGenome API, with a private preview rollout dedicated to commercial research. Scientists can access the tools to prioritize uncharacterized genetic variants linked to rare diseases and complex medical traits.