Google DeepMind announced AlphaGenome 2 on November 11, a DNA-analysis model that predicts how DNA changes may affect particular cell types and is available to researchers for noncommercial use.
Other models average activity across broad groups of tissues, but AlphaGenome 2 can make predictions for a selected cell type or state. This helps researchers examine whether the same DNA change acts differently in different cells. DeepMind says it trained the model on more than 100 million individual cell profiles from 68 tissues.
The model can also take a DNA sequence of up to one million letters along with information about a biological condition. In the company’s reported test using the TenK10K immune-cell dataset, AlphaGenome 2 identified 30% of DNA changes that caused an effect. Its previous model, which had been further trained on single-cell data, identified 16%.
Those results are reported by the company and do not show that the system can diagnose or treat disease. Researchers can access the model through the AlphaGenome API for noncommercial research. Access for commercial research is in a private preview.
DeepMind presents it as a tool to help scientists study how diseases work, not as a product for clinical use.