Additional file 2: Figure S1. DILI label prediction performance (balanced accuracy) of RF and SVM models trained using the DILIrank (−vLessConcern) dataset and Mordred molecular descriptors for 5-fold LOCO-CV, external test set, and FDA validation set (Methods). The balanced accuracy for 5-fold internal cross-validation, external test set, and FDA validation set for 10 models trained using different training data sets (DILIrank (−vLessConcern), DILIrank, DILIrank (+SIDER)) and training dataset splits is shown via whisker plots. The median model performance of 3 y-scrambled models is shown as triangles for cross-validation and external test set. Predictive accuracy is stable between cross-validation and external test set, but a distinct drop in predictive accuracy is observed when predicting the FDA validation set.