Skip to content
DraconiticResearch
Search Sources API
197,768 studies31 sourcesUpdated 55 min ago
PreprintCC BYbioRxiv

Benchmarking single-cell foundation models reveals an advantage in the low-data regime for cell-type and disease-state prediction

  • H. Benhayon1
  • M. M. Danziger
  • D. Ben-Zvi
  • M. Rosen-Zvi
  1. 1The Institute for Medical Research Israel-Canada, Department of Developmental Biology and Cancer Research, Jerusalem, Israel
Read full textPDFDOICite

Abstract

Single-cell foundation models perform well on cell-type classification, but their value for disease-state prediction and their advantage over classical representations remains unclear. We benchmarked cell-representations from four foundation models using logistic regression, on 868 disease-prediction tasks (82 datasets) and 65 cell-type datasets from CELLxGENE. With full data, the baseline (log-normalized counts) outperformed the best model (TranscriptFormer-Sapiens) for cell type and matched it for disease. However, with few labeled cells, TranscriptFormer-Sapiens significantly outperformed the baseline in both tasks up to ~102 samples per class. Our comprehensive zero-shot benchmark identifies the data-scarce regime as a domain where Single-cell foundation models truly show advantage.

Related

  1. Decision benchmarking reveals when protein foundation models add experimental value
    W. Ding, C. Lu, Y. Li +2
    bioRxivPreprintAll rights reservedFull textSystems & synthetic biologybioRxiv
  2. Harmonised benchmarking of foundation models for single-cell and spatial transcriptomics reveals context-dependent generalisation
    Sally Chen, Roxana Zahedi, Lucy Chhuo +12
    arXivPreprintCC BY-SA 4.0Full text1 citedCell biologyGenetics & genomicsarXiv
  3. Leakage-controlled benchmarking reveals generalization limits of deep learning for protein-ligand binding affinity prediction
    L. Wang, J. Cheng
    bioRxivPreprintCC BY-NC-NDFull textBioinformatics & computational biologybioRxiv
  4. Benchmarking single-cell foundation models for aging biology
    X. Ni, Y. Liang, J. Zhu +8
    bioRxivPreprintCC BYFull textBioinformatics & computational biologybioRxiv
  5. EEG-FM-Compass: progress, benchmarking and future directions for EEG foundation models
    Dingkun Liu, Yuheng Chen, Zhu Chen +5
    National Science ReviewPeer-reviewedCC BY31 citedNeuroscienceOpenAlex

Cite

BibTeXRIS
BibTeX
@misc{benhayon2026benchmarking,
  title = {{Benchmarking single-cell foundation models reveals an advantage in the low-data regime for cell-type and disease-state prediction}},
  author = {H. Benhayon and M. M. Danziger and D. Ben-Zvi and M. Rosen-Zvi},
  howpublished = {bioRxiv},
  year = {2026},
  month = oct,
  doi = {10.64898/2026.10.12.758421},
  url = {https://www.biorxiv.org/content/10.64898/2026.10.12.758421v1},
  note = {CC BY},
}
RIS
TY  - UNPB
TI  - Benchmarking single-cell foundation models reveals an advantage in the low-data regime for cell-type and disease-state prediction
AU  - H. Benhayon
AU  - M. M. Danziger
AU  - D. Ben-Zvi
AU  - M. Rosen-Zvi
PY  - 2026
DA  - 2026/10/12
PB  - bioRxiv
DO  - 10.64898/2026.10.12.758421
UR  - https://www.biorxiv.org/content/10.64898/2026.10.12.758421v1
AB  - Single-cell foundation models perform well on cell-type classification, but their value for disease-state prediction and their advantage over classical representations remains unclear. We benchmarked cell-representations from four foundation models using logistic regression, on 868 disease-prediction tasks (82 datasets) and 65 cell-type datasets from CELLxGENE. With full data, the baseline (log-normalized counts) outperformed the best model (TranscriptFormer-Sapiens) for cell type and matched it for disease. However, with few labeled cells, TranscriptFormer-Sapiens significantly outperformed the baseline in both tasks up to ~102 samples per class. Our comprehensive zero-shot benchmark identifies the data-scarce regime as a domain where Single-cell foundation models truly show advantage.
ER  -