Google DeepMind Unveils AlphaGenome 2 to Predict DNA Mutation Effects

What Happened

Google DeepMind has unveiled AlphaGenome 2, a new AI model designed to study how changes in DNA affect various cells [1]. This model is the successor to the previous AlphaGenome model released last year [1]. While most human cells share the same DNA, different cell types—such as skin, heart muscle, or immune cells—switch specific genes on and off depending on their activity [1]. AlphaGenome 2 is built to capture these differences by predicting the effects of genetic variants on individual cell types and states [1].

Key technical details include:

• The model was trained on more than 100 million individual cell profiles across 68 different tissues [1].
• According to Google, the model identifies nearly twice as many variants that change gene expression compared to its predecessor [1].
• Researchers found the model can predict outcomes for new cell types and biological states without requiring retraining [1].
• It can also identify gaps in existing experimental data to assist laboratory investigations [1].

Research scientist Adam Gayoso stated that the breakthrough involves tying human genetics to cell biology, allowing researchers to see how genetic variants affect cells at a granular level [1]. Žiga Avsec, the genomics lead at Google DeepMind, noted that this specificity is vital for drug development, as drugs target specific proteins and cell types [1]. Avsec also mentioned that while AlphaGenome 2 is a successor, it is not replacing the original model immediately because its detailed predictions are more computationally expensive to produce [1].

Availability and future plans:

• As of Wednesday, November 11, 2026, AlphaGenome 2 is available for non-commercial research via the AlphaGenome API [1].
• A limited private preview is offered for commercial research [1].
• DeepMind plans to improve genetics models by generating more data for underrepresented cell states [1].
• Future versions may be trained on data across multiple species [1].
• Long-term goals include using such models to create virtual cells, which are digital simulations of biological systems [1].

AlphaGenome 2 is part of a broader suite of biological AI models from DeepMind, which includes AlphaFold for protein structures, AlphaMissense for assessing mutations, and AlphaProteo for designing protein binders [2].