Under the strong alliance of artificial intelligence (AI) and neuroscience, the University of California, San Francisco and the Allen Institute team have jointly developed an AI model called CellTransformer, which helps to draw the most detailed mouse brain map currently available, including 1300 brain regions and subregions. This achievement reveals the structure of the brain with unprecedented precision, allowing scientists to correlate functions, behaviors, and disease states with smaller and more specific cellular regions, opening up new directions for exploring the workings of the brain. The relevant results were published in the latest issue of the journal Nature Communications. The core of the new model lies in its Transformer architecture, which is similar to the technical principles used in large models such as ChatGPT. Researchers say that the Transformer model is good at understanding contextual relationships. In the past, it was used to analyze the connections between words in sentences, while the CellTransformer was used to analyze the relationships between adjacent cells in space. It can predict the molecular characteristics of a cell based on its "neighborhood structure", thereby constructing a detailed brain tissue map. Unlike previous brain maps mainly based on cell types, the new findings focus on the structure of brain regions themselves. It relies entirely on data generation, with boundaries automatically defined by cellular and molecular features, rather than relying on manual experience judgment. With its precise division of 1300 brain regions and subregions, this map has become one of the most accurate and complex data-driven maps in animal brains to date. Research has shown that CellTransformer can not only accurately reproduce known brain regions such as the hippocampus, but also discover new and more detailed subregions in poorly understood areas such as the midbrain reticular nucleus. The team members vividly explained that it was like going from a map that only marked continents and countries to a map that could see cities. This new brain region division is entirely based on data rather than manual annotation, revealing a large number of unknown regions. Based on previous research, these newly discovered regions are likely to correspond to unexplored brain functions. The team points out that the potential of this AI model far exceeds that of neuroscience. The algorithm of CellTransformer has tissue universality and can be applied to other organ systems and even cancer tissues. With the help of spatial transcriptome data, it reveals the biological mechanisms in health and disease, providing new tools for drug development and disease treatment. (New Society)