# Universal Cell Embedding (UCE)

Source: https://onco.cc/technologies/universal-cell-embedding/  
OnCo record `universal-cell-embedding` (Technology). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

Universal Cell Embedding maps any cell from any species into one shared space without retraining.

## Summary

Universal Cell Embedding (UCE) is a transformer that represents each cell through ESM2 protein embeddings of the genes it expresses, so genes from different species map into the same space and no retraining is needed for a new organism. Developed at Stanford with the Chan Zuckerberg Initiative and described in a 2023 bioRxiv preprint, it was trained on 36M cells across 8 species and enables zero-shot cell type mapping, placing a new cell into a universal atlas without labels. It is useful for cross-species comparison, for example relating mouse tumour models to human samples. Its embeddings are coarse for fine perturbation effects, so it is a mapping tool rather than a model of how cells respond to drugs. For a newcomer: UCE puts every cell from any species onto one shared map.

## Fields

- Kind: Technology
- Status: emerging
- Last checked: 2026-09-08
- Tags: foundation-model; virtual-cell
- Principle: Transformer over ESM2 protein embeddings of expressed genes.
- Since: 2023
- Strengths: Cross-species zero-shot
- Limitations: Coarse for fine perturbation effects

## Sources

- bioRxiv 2023: https://www.biorxiv.org/content/10.1101/2023.11.28.568918v1

## Connected records

- fronts: [AI & Computation](https://onco.cc/fronts/ai-computation/)
- companies: [Chan Zuckerberg Initiative (Biohub)](https://onco.cc/companies/chan-zuckerberg-initiative/)
- institutions: [Stanford Health Care / Stanford Cancer Institute](https://onco.cc/institutions/stanford/)
- roadmaps: [Virtual cell roadmap: from bulk omics to a predictive model of a cancer cell](https://onco.cc/roadmaps/virtual-cell/)

---
JSON: https://onco.cc/api/v1/entities/universal-cell-embedding.json