For the perturbation generalization scenario, we curated 17 datasets that broadly cover two perturbation types, including (1) 13 genetic perturbation datasets, of which nine focus on single perturbations and four on combination perturbations; (2) 4 chemical perturbation datasets, including three single perturbation datasets and one combination perturbation dataset. In total, the perturbation counts in these datasets range from 25 to 1,618, ensuring broad applicability and robust evaluation of the methods.