Google DeepMind has introduced AlphaGenome 2. The AI model predicts how DNA changes affect gene activity in different cell types and states. Google hopes to gain insights into disease causes and potential targets for new drugs.
Almost all body cells possess the same DNA. Depending on their function, different genes are active within them. This applies, for example, to heart muscle cells or immune cells that fight infections. How small DNA changes affect gene activity in various cells has been difficult to predict so far.
AlphaGenome 2 examines DNA segments up to one million base pairs, the chemical building block pairs of genetic material. Researchers can computationally alter individual DNA elements and have the effects on gene activity of different cell types predicted.
New insights into gene activity
Google introduced the first version of AlphaGenome last year. In September, AlphaGenome Atlas followed, a database with predictions for approximately nine billion possible changes to individual DNA building blocks. The Atlas itself is not an AI model.
While AlphaGenome 1 creates predictions for many tissues and cell lines simultaneously, AlphaGenome 2 specifically considers individual cell types and their states. This allows it to detect differences that are hidden in average values of entire tissues. Google trained the model using data from humans and mice. To be able to adapt it to new cell states, the company additionally used 107 million single-cell profiles from 68 human tissues.
Researchers can also input their own measurement data on gene activity into the model. This allows it to investigate previously unknown cell states without retraining and predict biological properties that haven't been measured there yet.
Videos by heise
