Scientists have used AI models like ChatGpt to create new maps of the mouse brain that capture previously unknown regions of the organ in unprecedented detail.
The map, published on Tuesday (October 7th) in Journal Nature Communications, captures 1,300 areas of the brain, and is the first region to detail brain regions without the need for manual input from humans. Research authors at the University of California, San Francisco (UCSF) and the Allencel Science Institute hope that the project will allow researchers to sketch across such an organizational map.
You might like it
This approach also shows the location of individual cells within the space within tissues. This information was informed of previous cellular atlases in the mouse brain. However, placing information from such experiments on a comprehensive brain map presents an important challenge. For previous maps of the brain, researchers had to manually annotate each part of the map to distinguish between specific areas of the brain and where recorded cells would fit within them. New research avoided this troublesome task.
The spatial transcriptomic data used for the new map contained information on the activity of 500-1,000 genes in each cell analyzed. At this level of complexity, data analysis is challenging, said Reza Abbasi-Asl, professor of neurology and bioengineering at UCSF. Furthermore, Abbasi-ASL says that marking brain regions using raw spatial transcriptome data (a process called parcels) generates ambiguous maps.
That’s where the team’s AI-based approach paid off.
Large-scale language models (LLMs) such as ChatGpt attract and encourage millions of users with the ability to generate text output from prompts. At their core, these systems work by mathematically predicting the relationships between individual words. Abbasi-Asl, alongside doctoral student Alex Lee, created an AI system called CellTransformer. This instead analyzes how individual cells are adjacent in the brain based on spatial transcription tumor information.
AI systems transform spatial data and enhance it with new information. “We build the missing part between spatial transcriptome data and the splitting of the brain that connects the two,” Abbasi-ASL told Live Science. According to Abbasi-ASL, the new dataset generated by CellTransformer generates sharper maps that are more similar to known brain regions than manual annotations, identifying fine particle regions that were not previously inspired.
Image 1/2
The new map covers approximately 1,300 sections of the mouse brain, resulting in a total dataset of over 9 million cells. The team coordinated data with Allen Laboratory’s Common Coordinate Framework (CCF), a high-resolution map of the mouse brain previously constructed using manual annotations. There was a strong consistency between the AI-generated power and the gold standard CCF, giving the team confidence that their findings were very accurate.
CellTransformer successfully mapped known brain regions, such as the key memory center, hippocampus. The tool also struggled to retrieve data for other mapping efforts, such as the midbrain reticulomatous (CK), located at the top of the brainstem, charted areas of the brain while regulating sleep while processing sensory and motor information.
You might like it
The authors say that the data processing behind CellTransformer does not work alone in brain tissue.
“Similar pipelines can now be used in datasets emerging from the heart, from other body parts, and from tissues collected in disease models as opposed to health models,” Abbasi-ASL said.
The team also wants to test cell transformers on human brain data, but while the mouse brain contains millions of cells, our brains have around 170 billion cells, including 86 billion neurons. The pure size and more complex structure of the human brain make it difficult to provide enough spatial data to feed AI.
However, if such data can be brought to CellTransformer, Abbasi-ASL believes the tool can handle it. “We believe it works for human data too,” he said. “That’s another really important next step.”
Source link
