Supplementary Figure 11: Evaluation of cell type bias in target gene and ligand activity prediction performance of NicheNet.

To evaluate the possible presence of a cell type bias, we assessed the predictive performance of NicheNet on ligand treatment datasets that were only selected out of the total set of 111 datasets if they measure the response to a ligand for which the response was profiled in at least two different cell types that were stimulated by at least two different ligands. For the 43 selected ligand treatment expression datasets, we show the target gene (a-b) and ligand activity (c-d) prediction performances in function of the ligand added to the cells and the cell type that was stimulated. For each ligand-cell type combination, we also indicate the number of analyzed datasets. When multiple datasets were analyzed for a specific ligand-cell type combination, the average performance is shown on the heatmap. AUROC: area under the receiver operating characteristic curve; AUPR (corrected): area under the precision-recall curve, corrected for random prediction.