Supplementary Figure 7: Performance of several feature importance scores in predicting ligand activity.

For 111 ligand treatment expression datasets, we assessed the ability of NicheNet in predicting with which ligand cells were treated based on the set of genes that are differentially expressed upon ligand treatment. For this procedure, ligand-target regulatory potential scores were used to calculate feature importance measures for all ligands in every expression dataset separately (n = 111 datasets; 51 unique ligands) and these importance measures were in a next step used to predict the ligand activity state of each ligand in each dataset. The performance of each feature importance measure in ligand activity prediction (as evaluated by the AUROC and AUPR) is shown for different feature importance measures: spearman: Spearman’s rank correlation; pearson: Pearson correlation; mean_rank_GST_log_pval : negative natural logarithm of the mean-rank gene-set enrichment p-value; auc_iregulon_corrected: area under the cumulative recovery curve (considering the top 3% of the ranking), corrected for random prediction; aupr_corrected: area under the precision-recall curve, corrected for random prediction; auroc: area under the receiver operating characteristic curve. Boxplot elements: center line: median; box limits: upper and lower quartiles; whiskers: 1.5x interquartile range; points: outliers. The violin plots show the probability density of the data (tails of the violins are trimmed to the range of the data). Red dashed line: performance of random guessing.