Figure 4

Model performance on the test set. RMSE_test (upper panel) and \(R^{2}_{0\ test}\) (lower panel) values for the following models: (group A) single PCM, (group B) Family QSAR and Family QSAM, (group C) individual QSAR, (group D) model ensembles comprising those single PCM models exhibiting the highest predictive power, and (group E) model ensembles comprising the whole model library. Bars are colored according to the groups defined in Table 2. Confidence intervals correspond to the mean value +/- one standard deviation calculated with bootstrapping [74].