Fig. 8.

Performance of algorithms to predict DTIs against unseen data. Receiver Operating Characteristic (ROC) plots showing the true positive rate (TPR) against the false positive rate (FPR) are calculated for machine learning models pre-trained on the GraphDTI dataset and applied to classify unseen instances from the PubChem BioAssay dataset. The performance of several DTI predictors is presented, GraphDTI (solid blue lines), EnsemDT (dashed pink lines), EnsemKRR (dashed-dotted green lines), and RLS-Kron (dotted red lines). The gray area corresponds to the performance of a random classifier