Universal cell embedding model captures cell diversity across species and tissues

A new Universal Cell Embedding (UCE) foundation model can represent cells across diverse species and tissues. UCE was trained on 36 million cells using self-supervision, creating a unified biological space that captures cellular variation and noise. It allows embedding new cells without retraining. The model exhibits emergent behavior, identifying biological patterns not explicitly trained for, and aids in analysis, annotation, and hypothesis generation for single-cell data.