Fig. 5
From: On the robustness of generalization of drug–drug interaction models
Impact of evaluation schemes on the different categories of phenotypes. The phenotypes are grouped according to their frequencies in the datasets. For example, [11,22[ in x-axis groups all phenotypes present at greater than 11% and less than 22% in the training set. On the other hand, for DrugBank, there are only 3 different intervals because there is no phenotype present above 25%, which is the case in Twosides and Twosides-NOSYN. AUPRC and AUROC scores are from our best model (1D-CNN) on the benchmark datasets