Google DeepMind has introduced AlphaGenome 2, an AI model designed to predict how genetic variants influence gene expression across different cell types and states. This new iteration succeeds the original AlphaGenome model released last year, offering a more granular analysis of cellular behavior.
The model was trained on over 100 million individual cell profiles spanning 68 distinct tissues. According to DeepMind, AlphaGenome 2 identifies nearly twice as many variants affecting gene expression compared to its predecessor and can make predictions for new cell types without requiring retraining.
The increased precision of AlphaGenome 2 comes with higher computational costs, making it more resource-intensive than the previous version. Consequently, DeepMind does not intend to replace the original model, as the new version is less practical for large-scale, genome-wide mapping tasks.
Google has made AlphaGenome 2 available for non-commercial research through the AlphaGenome API, with a private preview currently offered for commercial entities. The company plans to continue improving its genetics models by incorporating more diverse data, including information from multiple species.
