Supplementary Figure 10: The contribution of individual data sources to target gene and ligand activity prediction performance can vary over different ligand treatment datasets.

To assess the effect of individual data sources on target gene and ligand activity performance prediction performance on individual ligand treatment datasets, we evaluated the performance of ‘leave-one-in’ models of ligand-target regulatory potential scores in which every data source in a specific layer (ligand-signaling or gene regulatory) was left out, except an individual data source of interest. Hereby, differences in model performances between leave-one-in models of the same layer can be attributed to differences in information content between the corresponding data sources. It should be remarked that a distinction between ligand-receptor and signaling interaction data sources was in this figure only made for visualization purposes, not for the construction of leave-one-in models. For two pairs of ligand treatment datasets, we compared in (a) the target gene prediction performances and in (b) the ligand activity prediction performances of the individual ligand-signaling (red and dark blue) and gene regulatory (orange) leave-one-in models to each other and to the complete final model (light blue). AUROC: area under the receiver operating characteristic curve. Red dashed line: performance of random guessing.