CellBench-LS
Rigorous framework for evaluating generalization of single-cell foundation models in low-supervision scenarios across clustering, batch correction, annotation, expression reconstruction, and perturbation prediction.
Composite
65.2
Experimental validation
Retrospective
Stages
Virtual CellDisease Modeling
Modalities
small-moleculebiologic
Task types
classificationregressiongeneration
Size
tasks: 5
datasets: multiple general + multi-batch + perturbation scRNA-seq
models_evaluated: 10
datasets: multiple general + multi-batch + perturbation scRNA-seq
models_evaluated: 10
License
Other
First release
2026-04-01
Last updated
2026-04-01
Official site
Leaderboard
→ leaderboard
Dataset
→ dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
CellBench-LS: Benchmarking Single-Cell Foundation Models in Low-Supervision Scenarios · Unknown — biorxiv preprint · 2026 · paper · doi:10.64898/2026.04.01.714123 · 0 citations
Flags
none
Experts
—
Groups
—
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
4
coverage
4
maintenance
3
adoption
2
quality
4
accessibility
3
industry_relevance
3
Notes
Addresses critical gap: most scFM benchmarks use full supervision. Low-supervision evaluation is more realistic for clinical/translational settings. Finds scFMs don't consistently outperform simpler baselines.