Google DeepMind has unveiled AlphaGenome 2, a new artificial intelligence model that can predict how even the smallest changes in DNA, such as replacing one of its components, will affect the functioning of different human cells. The development should help scientists better understand the causes of diseases and find possible directions for creating drugs.
Google DeepMind has unveiled AlphaGenome 2, a new artificial intelligence model that can predict how even the smallest changes in DNA, such as replacing one of its components, will affect the functioning of different human cells. The development should help scientists better understand the causes of diseases and find possible directions for creating drugs.
Google reports this.
All the cells in our body have almost the same set of DNA, but they use it differently. For example, heart cells need some genes, while immune system cells need others. Because of this, even a small change in DNA can have different consequences depending on which cell it acts on. It is these differences that AlphaGenome 2 will try to predict.
The previous version of the model analyzed the impact of DNA changes mainly at the level of whole tissues. The new one can look at individual cell types and predict which genes will become more active and which will become less active.
This has already been tested in real-world disease models. For example, researchers previously found that a certain DNA change contributes to the development of inflammatory bowel disease by overactivating the ETS2 gene in certain immune cells.
AlphaGenome 2 was able to replicate this discovery computationally. The model determined which DNA change affected the gene and in which cells it occurred. Its prediction showed that the change affected only some immune cells in the gut the most, while other cells were barely affected.
In practice, this could help scientists find the causes of diseases and potential targets for new drugs more quickly. Instead of testing a multitude of genetic changes in different cells, researchers can first focus on the ones that the AI thinks are most important.
And AlphaGenome 2 can predict the impact of DNA changes even in cells whose state it has not previously studied. For example, after vaccination, when the work of immune cells changes. To do this, scientists only need to provide the model with new data on gene activity in a few dozen cells. This is important, because some of the consequences of genetic changes may not be noticed if cells are studied only in their normal state.
Google used data from more than 107 million individual human and mouse cells to train AlphaGenome 2. Scientists can now use the model for non-commercial research via the AlphaGenome API. It is available for commercial research in a closed beta.
Previously, dev.ua wrote that Anthropic's Claude helped discover a new biological system similar to CRISPR, a technology used to edit DNA. To do this, about 950 AI agents analyzed genetic data for 21 hours.
