Fig. 3. Evaluation of our method predictive performance on identifying the seriousness of clinical outcomes of adverse reactions to drugs.

a and b show the distributions of AUCs and AUPRs on the task of identifying whether the adverse reactions to drugs cause serious clinical outcomes and the task of predicting the corresponding classes of serious clinical outcomes caused by seirous adverse reactions to drugs, respectively. The results are obtained by tenfold cross-validations based on the benchmark dataset. c depicts the Receiver Operating Characteristic (ROC) and Precision-Recall (PR) curves on the task of identifying whether the adverse reactions to drugs cause serious clinical outcomes based on two independent datasets, respectively. d shows the distributions of AUCs and AUPRs in the task of predicting the corresponding classes of serious clinical outcomes caused by serious adverse reactions to drugs.