CellPB is a benchmarking framework with rich collection of data to systematically evaluate in silico perturbation methods for 4 key scenarios in terms of both their performances and usability:
- (1) Predicting effects of previously unobserved perturbations in known cell types,
- (2) Predicting the effects with known perturbations in previously unobserved cell types,
- (3) Transferring predictions to bulk RNA-seq data of cell lines without prior training on them, and
- (4) Predicting effects of key genes driving cell state transitions in specific biological processes. By clearly distinguishing these scenarios, the framework offers a flexible and fair evaluation of various computational methods.
