PerturbArena
Comprehensive benchmark for comparing single-cell perturbation prediction models. 12 models + 3 baselines across 25 datasets with 24 metrics. Three core tasks: unseen perturbations, combinatorial perturbations, and cell state transfer across conditions.
Composite
74.4
Experimental validation
None
Stages
Target ID
Modalities
chemical_perturbationgenetic_perturbationsingle_cell
Task types
perturbation_predictioncombinatorial_predictioncell_state_transfer
Size
models: 12
datasets: 25
metrics: 24
datasets: 25
metrics: 24
License
Unknown
First release
2024-12
Last updated
2025-06
Official site
Leaderboard
→ leaderboard
Dataset
→ dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
PerturbArena: A Comprehensive Benchmark for Single-Cell Perturbation Prediction · · 2024 · paper · doi:10.1101/2024.12.23.630036 · 12 citations
Flags
virtual_cellmulti_metric
Experts
—
Groups
—
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
4
coverage
4
maintenance
3
adoption
3
quality
4
accessibility
4
industry_relevance
4
Notes
Complementary to scPerturBench. Emphasizes metric divergence analysis and practical method selection guidelines. Shows limited robustness to shifts across cellular contexts. Chinese research group.