Fig. 2.

Model performances of each DDI prediction case. The model performances are the average values of all side effect results. We compared our models to Decagon (orange) and the ANN model; from Shankar et al. (green). A Our model showed good performances compared to previous studies. The performance of ANN model is implemented results with our dataset, while the reported performances in the original paper are AUC of 0.634, accuracy of 0.787, and F1 of 0.631 obtained by averaging prediction performances of 243 side effects. B, C Decagon was not applicable to predict unseen drugs, and the ANN model was implemented