PerturbArena

A Systematic Comparison of Single-Cell Perturbation Response Prediction Models

Comprehensive benchmarking of 12 models + 3 baselines across 25 datasets with 24 evaluation metrics in 3 core prediction tasks.

Study Overview

Figure 1: Study overview showing datasets, perturbation types, and evaluation framework

Benchmarking Tasks

1
Unseen Perturbations

Predict cellular responses to perturbations not seen during training, testing model generalization.

View Results
2
Combinatorial Perturbations

Predict responses to combinations of perturbations, including interaction effects and synergy.

View Results
3
Cell State Transfer

Transfer learned perturbation effects across different cell types and experimental conditions.

View Results

Research Impact

This comprehensive evaluation provides the first systematic comparison of perturbation prediction models, establishing performance baselines and identifying key factors affecting model success in single-cell biology applications.

25

Datasets

15

Models Evaluated

24

Evaluation Metrics