A Systematic Comparison of Single-Cell Perturbation Response Prediction Models

Predicting single-cell transcriptional responses to perturbations is central to dissecting gene regulation and accelerating therapeutic design, yet the field lacks a rigorous, task-spanning assessment of model behavior. We present a large-scale benchmark of 12 representative methods and 3 baselines across 25 datasets spanning diverse perturbation modalities and species, including two new primary immune-cell drug-response resources. We evaluated three core tasks--generalization to unseen single-gene perturbations, prediction of combinatorial interactions, and transfer across cell types--using 24 metrics covering expression-level accuracy, relative changes, differential expression recovery, and distributional similarity. Across tasks, performance depended strongly on perturbation effect size and evaluation perspective: expression-level agreement was highest for small-effect perturbations resembling controls, whereas delta- and DE-based metrics improved with larger effects, providing clearer signals. Models shared a conservative bias, with fine-tuned foundation models compressing variance and underestimating synergistic effects in combinations. PerturbNet showed superior recovery of DE signatures in Tasks 1 and 2, while no method consistently generalized across cell types in Task 3, where dataset heterogeneity dominated outcomes. This benchmark establishes current methodological limits, clarifies when metrics diverge, and provides a foundation for developing virtual-cell models that more faithfully capture heterogeneous perturbation responses.