基础模型如何比较跨物种、组织与实验的数据
Single-cell experiments generate detailed molecular portraits, but their coordinates are not naturally shared. A liver cell measured in one laboratory may be difficult to compare with a related cell from another tissue, species or technology.
A foundation model seeks to learn a common representation from a very large collection. In the reported work, training data covered 36 million cells, hundreds of experiments, dozens of tissues and eight species, allowing cellular similarities and differences to be encoded in a reusable numerical space.
Inspired by a public Nature title/summary · Rewritten by Vocabsavvy · Vocabsavvy Original (public-snippet inspired)