Neuroscientists have turned to artificial intelligence to create the most detailed maps of the brain yet, uncovering previously hidden neighborhoods of cells that could reshape our understanding of brain function. At the heart of this breakthrough is a machine learning tool called CellTransformer, developed by researchers at the Allen Institute for Brain Science and the University of California, San Francisco. The algorithm analyzed genetic data from over 10 million individual cells across five mouse brains, identifying hundreds of distinct cellular neighborhoods that were invisible to traditional mapping methods. For decades, brain maps were based on visual differences in cell density and arrangement, like the famous Brodmann areas created in the early 1900s. While these maps were foundational, they were limited by human subjectivity and couldn’t capture the complexity of cellular diversity. With advances in single-cell RNA sequencing, scientists can now measure which genes are active in each cell. This has revealed thousands of unique cell types—more than ever before. But simply knowing the types of cells isn’t enough. To understand how the brain works, researchers need to know how these cells are organized in space. The key insight came from Bosiljka Tasic and Reza Abbasi-Asl, who realized that the brain’s structure emerges not from individual cells, but from the patterns of how different cell types cluster together. CellTransformer was designed to mimic this logic. It learned to predict a cell’s identity based on its neighbors, gradually building a high-resolution 3D map of cellular neighborhoods. The algorithm didn’t just replicate existing maps—it discovered new subdivisions in regions long thought to be uniform. One striking example is the caudoputamen, a brain structure involved in movement, reward, and decision-making. Traditional maps show it as a single, large region. But CellTransformer revealed it’s actually made up of multiple smaller, distinct neighborhoods. These subregions align with earlier findings based on neural connections, suggesting they may have specialized roles. This could explain why scientists have long debated the caudoputamen’s functions—because they were looking at the wrong level of detail. The algorithm also uncovered four new neighborhoods in the midbrain reticular nucleus, a region critical for initiating movement. Each of these areas had unique combinations of cell types and gene activity, including some that had previously been classified as belonging to entirely different brain regions. While the maps are still being validated, the results are promising. The algorithm matched known brain structures with high accuracy and consistently found new ones across different mouse brains, regardless of sex or slicing direction. This consistency suggests the findings are biologically meaningful, not artifacts. Experts like Claudia Doege and Hourig Hintiryan see this as a major leap forward. “We can now see things that a human eye cannot,” Tasic said. The next step is to link these neighborhoods to specific functions—by manipulating them in animal models and observing changes in behavior. The ultimate goal is to apply the same method to human brains. Though human data is still limited, researchers believe CellTransformer can handle the scale. When enough data becomes available, it could reveal new insights into brain disorders like Alzheimer’s, schizophrenia, and Parkinson’s, where disruptions in cellular organization may play a role. Beyond the brain, the algorithm could map other organs, helping compare healthy and diseased tissues. As Yongsoo Kim put it, “AI is a helper for the human.” With tools like CellTransformer, neuroscience is entering a new era of discovery—where the map itself becomes a guide to understanding the mind.