🇺🇸▼
Real estate agents always say the most important rule for buying a home is location. They repeat the phrase, "location, location, location." In the field of neuroscience, this rule is even more critical. Bosiljka Tasic, a researcher who calls herself a "biological cartographer," explains that location is everything in the brain. A small injury in one spot could destroy a person's memory, while damage in another place might change their personality completely. Without a precise and detailed map of these internal areas, neuroscientists and doctors are effectively lost.
Scientists have tried to map the human brain for over a hundred years. Early researchers traced patterns of cells that were only visible under a microscope. They made colorful charts to show different regions and what they did. Today, the level of detail is much higher. Scientists can now look at the brain cell by cell, looking at the genetic activity inside each one. Despite these advances, the maps often look incomplete or inconsistent. Some large brain regions do many different tasks, so scientists suspect they should be split into smaller, specialized areas. Mapping these cellular neighborhoods from huge genetic datasets has been a difficult and time-consuming task.
Recently, Tasic, a neuroscientist at the Allen Institute for Brain Science, and her team asked artificial intelligence to help. They fed genetic data from five mouse brains into a special computer program. The data included 10.4 million individual cells, with hundreds of genes in each cell. The program generated maps that showed both known and new subdivisions within larger brain regions. Humans could not draw these complex borders in many lifetimes, but the computer finished in just hours. The team published their methods in the journal Nature Communications in October.
By using this method on other animals and eventually humans, researchers hope to show the brain's finer layout. They also want to create new ideas about how different parts of the organ work in health and disease. "We want to understand how the cells are organized in three-dimensional space," said Claudia Doege, a neuroscientist at Columbia University who was not part of the study. "Only if we know how they are organized can we figure out how they can potentially work with each other."
Brain mapping is an old science. It dates back to the early 1900s when Korbinian Brodmann, a German neuroscientist, defined regions of the cerebral cortex. This is the outer, thinking part of the brain. Brodmann stained brain slices with dye and studied them under a microscope. He traced lines to create a map of 52 regions known as Brodmann areas. For decades, scientists used similar methods. "What anatomists used to do is, they have a pencil, and they draw the line" between different-looking regions, said Yongsoo Kim, a neuroanatomist at Penn State College of Medicine. These maps are useful, but they are often subjective. When Kim asked senior scientists about their methods, the answer was often, "It's all in my head."
Recently, advanced molecular techniques allowed researchers to investigate individual cells. A cell's identity is determined by which of its tens of thousands of genes are turned on. Scientists can slice up a brain and measure the RNA molecules from each cell. RNA are copies of active DNA. They can map these genetic patterns back to the cells' original locations. This approach has distinguished thousands of individual brain cell types. The Allen Institute's latest mouse brain atlas, published in 2023, includes more than 5,000 different cell types.
However, these massive datasets did not produce the kind of brain map Tasic needed. The maps she created had regions that were not always "biologically meaningful," she noted. Most brain regions are not defined by a single cell type. Many cell types are not limited to one region. Instead, each area contains a mixture of cell types.
To understand the difficulty, imagine an airplane passenger looking out the window at a city below. If they try to identify neighborhood boundaries by looking at one building at a time, they cannot see the surroundings. To find neighborhoods, they need to see how different building types group together. To map the brain's subregions, Tasic needed to analyze how different cell types grouped together. This was not something her human brain could do alone with the RNA data. She needed better computer tools and a research partner.
Tasic found the perfect collaborator in Reza Abbasi-Asl, a computational neuroscientist at the University of California, San Francisco. "I have always been fascinated and intrigued by how we can leverage AI to understand cellular organization in the brain," he said. To define cellular neighborhoods, Abbasi-Asl and his graduate student Alex Lee started with RNA profiles from 3.9 million cells in a single mouse brain. They programmed a machine learning algorithm to choose one cell and hide its identity. Then the AI, called CellTransformer, would predict that cell's gene expression and type based on its neighbors. It would check if it was right and update its algorithm. By repeating this process millions of times, the algorithm learned how different types of brain cells group together. From there, it could build a high-resolution map of those groups.
Returning to the airplane analogy, what CellTransformer does is like holding up a thumb to the window to block one building and then predicting its type. The surroundings give clues about what kind of structure fits into the neighborhood. Approaching brain mapping as relationships between nearby cells was the "secret sauce," Abbasi-Asl said. It allowed the algorithm to map meaningful neural neighborhoods, each made of a blend of cell types. Depending on the detail requested, it could define anywhere from 25 to 1,300 neighborhoods in the mouse brain. With AI, "we see things that a human eye cannot see," Tasic said.
Using data from four additional mouse brains, CellTransformer produced similar maps. Doege said this is excellent evidence that the technique is reliable. The algorithm was not generating wholly new maps from scratch, so it could not "hallucinate" as some generative AI models can. Still, the team compared CellTransformer's output to known brain maps. They used the hand-drawn Allen Mouse Brain Common Coordinate Framework as a trusted comparator. The CellTransformer map was a good match, laying out similar structures such as the layers in the cortex.
The algorithm also identified new neighborhoods that previous methods had missed. Take the striatum, a striped structure near the middle of the brain. In maps of the mouse brain, this area is called the caudoputamen. "You just see one huge structure," said Hourig Hintiryan, a neuroanatomist at the University of California, Los Angeles. It is known to participate in movement, reward, and overall brain management. How could one piece of brain perform such different tasks? CellTransformer's explanation is that it is not one uniform brain region. The map confirmed the caudoputamen is subdivided into smaller areas, though researchers have not yet matched each region to a function. Moreover, the new subdivisions corresponded nicely to a map Hintiryan and colleagues published in 2016 based on a different technique.
Identifying such subregions across the brain, Hintiryan said, could resolve debates between neuroscientists who assign vastly different functions to the same large brain region. It seems likely that "they're both correct, they're just looking at different areas," she said. Abbasi-Asl and Tasic were thrilled that CellTransformer accurately matched known brain cartography and mapped novel subdivisions. For example, the brainstem's midbrain reticular nucleus, involved in initiating movement, is a fairly underexplored region. CellTransformer picked out four new neighborhoods there.
The Nature Communications paper serves mainly to introduce the CellTransformer method and show that it can find novel regions. The thousand-plus new neighborhoods still require validation. Drawing the map is just the beginning. What is most exciting is what scientists may be able to do with it. "The more granular our understanding of structure, the more specific we can get with our interrogations and interventions," Hintiryan said.
Emerging questions center on the functions of all these neural neighborhoods. To pinpoint what each bit does, scientists could eliminate or activate these newly identified regions in lab animals and check for behavioral changes. The real prize will be to apply CellTransformer to human brains. Doege suspects some neighborhoods will match well between mice and people, while others will diverge. Unfortunately, the quantity of data the algorithm needs is not available from human brains yet. While the mouse brain contains about 100 million cells, the human brain has around 170 billion, and that menagerie is still undergoing genetic analysis. When sufficient data becomes available, Abbasi-Asl and Tasic think CellTransformer will be up to the challenge.
They are also interested in incorporating other technologies, such as connection tracing, into CellTransformer. This would be like adding streets and highways to the city neighborhoods. Beyond the brain, the same algorithm could offer detailed cell maps of other organs. This would allow scientists to compare healthy versus diseased tissue. Human scientists simply cannot sort out these details on their own. "I see AI as kind of a helper for the human," Kim said. "Discovery will be accelerated in a dramatic way."