Fig. 1

Ligands are classified according to similarities in multidimensional signaling profiles using pharmacodynamic parameters. 320 virtual compounds were defined by logistic and operational parameters to represent 16 distinct signaling profiles describing response at six different readouts. Indicated parameters were then subject to NNMF followed by k-means clustering, to produce corresponding similarity matrices that were represented as heatmaps and hierarchical clustering trees (using the R heatmap function with the metric:ward.D2) (a–c) or t-SNE plots (using the R package tsne with default parameters) (d–f). Ligands were color-coded to highlight how different combinations of parameters differentiated compounds originating from the different profiles originally defined