Bright key facts / Demonstrated
One coordinate system for cells across species.
Universal Cell Embedding was trained across 36 million cells, more than 1,000 cell types, and eight species to create reusable representations for cell biology.
- AI’s role
- The model learned a common numerical representation of cells from large single-cell datasets.
- Documented result
- The Nature paper reports a corpus of 36 million cells spanning more than 1,000 cell types and eight species, with benchmarked downstream analyses detailed in the paper.
- Important limitation
- An embedding compresses data and can hide important biological variation.