OP3: single-cell multimodal dataset in PBMCs for perturbation prediction benchmarking

The Open Problems Perurbation Prediction (OP3) dataset is to-date the largest single-cell small molecule perturbation dataset in primary tissue with multiple donor replicates. Compounds are dosed in cultured peripheral blood mononuclear cells (PBMCs), enabling the robust computation of perturbation effects on gene expression across four distinct cell type populations: T-cells, B-cells, NK cells, and Myeloid cells. Each compound was also dosed in triplicate across three donors, two male and one female. We complemented this data by also profiling PBMCs from the same three donors at baseline with joint multimodal snRNA-seq/scATAC-seq. This dataset is the basis for the OP3 benchmark, which evaluates machine learning models on the task of extrapolation perturbation effects into unseen cell types.