人类画了100年的脑图，AI仅用几小时，还绘制出新脑区

Core Insights - A new machine learning algorithm called CellTransformer has been developed by a neuroscience team from the University of California, San Francisco, which can classify and map the brain structures of five mice in just a few hours [1][3][10]. Group 1: Technology and Methodology - CellTransformer utilizes an encoder-decoder architecture that predicts the identity and gene expression of cells by observing neighboring cells, significantly reducing the time and complexity involved in traditional brain mapping methods [7][9]. - The algorithm employs a self-supervised learning approach, where it randomly masks a cell's gene expression and predicts it based on the surrounding cells, allowing for effective interaction and learning of cellular relationships [9][10]. - The model can define between 25 to 1300 neural regions in the mouse brain without using brain region labels, achieving high alignment with known anatomical structures [10][14]. Group 2: Results and Findings - CellTransformer has demonstrated the ability to identify and delineate new brain regions that were previously unmarked by conventional neuroscience methods, such as the striatum, which is involved in movement and reward [14][16]. - The algorithm has also revealed that the striatum is not a single brain region but is subdivided into smaller areas, each potentially serving different functions, which opens new avenues for research [16][17]. - In addition to mapping mouse brains, CellTransformer has shown potential for application in other animals and could eventually be adapted for human brain mapping, although challenges remain due to the complexity and size of human brain data [19][21]. Group 3: Future Applications - The ultimate goal of CellTransformer is to be applied to human brain mapping, with researchers optimistic about its potential once sufficient data is available [21]. - Beyond brain mapping, the algorithm could also be utilized in other organs, such as kidneys, to provide detailed cellular maps that could aid in understanding health conditions like diabetes [21].