0
0The paper introduce Universal Cell Embedding (UCE), a groundbreaking foundation model designed to map the vast molecular diversity of biology into a single digital space. By leveraging a protein language model and transformer architecture, UCE represents cells across different tissues and species without requiring manual data labeling or specific model retraining. This "zero-shot" capability allows researchers to analyze new datasets—including those from species never seen by the model—while effectively correcting for experimental noise and batch effects. The researchers demonstrated the model’s power by creating the Integrated Mega-scale Atlas, which organizes 36 million cells into a unified landscape of over 1,000 cell types. Ultimately, UCE serves as a robust tool for cell type annotation, developmental lineage tracing, and cross-species discovery in modern genomics.
References:
Rosen Y, Roohani Y, Agrawal A, et al. Universal cell embedding provides a foundation model for cell biology[J]. Nature, 2026: 1-9.
