scPerturBench

Comprehensive benchmark of 27 single-cell perturbation response prediction methods across 29 datasets. Evaluates generalization to unseen perturbations, combinatorial interactions, and cross-cell-type transfer. Published in Nature Methods.

Composite

91.9

Experimental validation

Retrospective

Stages

Target ID

Modalities

chemical_perturbationgenetic_perturbationsingle_cell

Task types

perturbation_predictiongeneralizationmethod_comparison

Size

methods: 27
datasets: 29

License

MIT

First release

2025-12

Last updated

2025-12

Official site

→ project page

Leaderboard

→ leaderboard

Dataset

→ dataset

Code / GitHub

→ repository

HuggingFace

→ HF

Paper

Benchmarking algorithms for generalizable single-cell perturbation response prediction · · 2025 · paper · doi:10.1038/s41592-025-02980-0 · 15 citations

Flags

nature_methodscomprehensive_comparison

Experts

—

Groups

—

Hosted by

—

Related benchmarks

PerturbBench, Open Problems: Perturbation Prediction, X-Atlas/Orion (Xaira Genome-wide Perturb-seq)

Rubric (7-criterion)

rigor

coverage

maintenance

adoption

quality

accessibility

industry_relevance

Notes

Published in Nature Methods (Vol 23, Issue 2). Most comprehensive evaluation of perturbation prediction methods. Covers both genetic and chemical perturbations. Web interface for exploring results. Chinese research group (Tongji University).

← Back to all benchmarks

Compare:

Open comparison →