Figure 2.

Benchmarking pipeline for evaluation and comparison of CCC inference methods. For benchmarking 18 LR inference methods, we used 11 scRNA-seq data sets as input and evaluated them from three aspects: similarity of the predicted LR interactions by each pair of methods, the differential LR correlations (DLRC) in close and distant cell pairs in 11 ST data sets, and the prediction accuracy (AUPRC) based on cell type–specific CAGE and proteomics data. We also calculated the Jaccard index to evaluate the robustness of each method to different sampling rates of cells in 14 scRNA-seq data sets. For benchmarking five L/R-target inference methods, we used eight ST data sets as input and evaluated the prediction accuracy (AUPRC) using 15 sets of cell line perturbation-expression data.