Scientists at UC San Francisco and the Allen Institute have developed an artificial intelligence (AI) model that produced a detailed map of the mouse brain. This map shows 1,300 regions and subregions, including some that were not known before. The work helps connect specific brain parts to functions, behaviors, and health problems.
The map uses data from spatial transcriptomics, a method that shows where genes are active in different parts of tissue. This allows for a clear view of cell positions and activities without relying on human judgments.
The AI model, called CellTransformer, is key to this advance. It works like the technology in tools such as ChatGPT, using a transformer framework. Here, it looks at how cells near each other in the brain share features. By analyzing large sets of data, CellTransformer finds important subregions automatically. For example, it confirmed known areas like the hippocampus, a brain part involved in memory, and found new ones in less-studied spots like the midbrain reticular nucleus, which helps with movement.
How the map stands out
This brain map differs from older ones because it focuses on regions based on cell types and is fully driven by data, not expert opinions. It has more detail than most maps for any animal. To check its accuracy, the scientists compared it to the Common Coordinate Framework from the Allen Institute, a standard map used in brain research. The results matched well, giving confidence in the new findings. CellTransformer can also work on other body tissues, not just the brain. For instance, it might help study cancer by mapping cell patterns in tumors. This could lead to better ways to understand diseases and find treatments. Overall, the map opens doors for new experiments and ideas in science.
The scientists have described the methods and results of this study in a paper published in Nature Communications.