# Single-cell and transcriptome foundation models: UCE, GeneCompass, BulkFormer, BulkRNABert

Source: https://onco.cc/terms/single-cell-foundation-models/  
OnCo record `single-cell-foundation-models` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

These are transformer models pretrained on expression profiles: UCE and GeneCompass on tens of millions of single cells, BulkFormer and BulkRNABert on bulk tumour and tissue transcriptomes.

## Summary

Universal Cell Embeddings (Rosen and colleagues) is a 650-million-parameter cell model that represents genes by their ESM2 protein embeddings so it works across species; GeneCompass (Yang and colleagues) is a knowledge-informed model trained on over a hundred million human and mouse cells. Bulk-native models are fewer: BulkFormer (Kang and colleagues, Cell Systems 2026) is a 147-million-parameter model trained on about half a million bulk profiles and released with code and weights; BulkRNABert (Gélard and colleagues, InstaDeep) is a language model for bulk RNA-seq cancer prognosis under a research-only licence. Frozen bulk embeddings have not consistently beaten a genome-wide linear baseline on TCGA survival, so their value is still being established.

## Fields

- Kind: Term
- Last checked: 2026-09-24
- Also known as: BulkFormer; BulkRNABert; GeneCompass; UCE model; Universal Cell Embeddings; scFoundation model; transcriptome foundation model; bulk transcriptome model
- Tags: cansim-terms

## Notes

- Listed in the CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme; CanSim page path /terms/bulkformer.

## Sources

- Rosen et al., Universal Cell Embeddings: a foundation model for cell biology (bioRxiv 2023): https://doi.org/10.1101/2023.11.28.568918
- Yang et al., GeneCompass (Cell Research 2024): https://doi.org/10.1038/s41422-024-01034-y
- Kang et al., BulkFormer: a large-scale foundation model for bulk transcriptomes (Cell Systems 2026): https://doi.org/10.1016/j.cels.2026.101657
- Gélard et al., BulkRNABert: cancer prognosis from bulk RNA-seq based language models (bioRxiv 2024): https://doi.org/10.1101/2024.06.18.599483
- BulkRNABert code (InstaDeep): https://github.com/instadeepai/multiomics-open-research

## Connected records

- collections: [CZ CELLxGENE / Human Cell Atlas](https://onco.cc/collections/cellxgene-hca/)
- terms: [Cancer AI vocabulary (CanSim terms map)](https://onco.cc/terms/cancer-ai-vocabulary/), [Fine-tuning versus frozen features, and LoRA](https://onco.cc/terms/fine-tuning-vs-frozen/), [Foundation model](https://onco.cc/terms/foundation-model/), [Single-cell RNA sequencing (scRNA-seq, 10x Chromium)](https://onco.cc/terms/single-cell-rna-seq/), [Tokenisation (genes, tiles and sequence as tokens)](https://onco.cc/terms/tokenisation/)
- technologies: [Nicheformer (spatial single-cell)](https://onco.cc/technologies/nicheformer/), [scFoundation (BioMap)](https://onco.cc/technologies/scfoundation/)

---
JSON: https://onco.cc/api/v1/entities/single-cell-foundation-models.json