Comparison of prediction accuracies of three drug similarities in predicting drug-target interactions.

AUC values are presented when two prediction methods and two drug-target interaction (DTI) data sets are used. CH and SE indicate the drug similarity based on the chemical structure and side effect, respectively. CS and CSD indicate the drug similarity by combining CH and SE, and combining CH, SE, and DDI, respectively. The last column indicates that the kernel fusion method developed in Lanckriet et al.[20] is used for combining multiple kernels in SVM.*indicates the highest value for each combination of method and data source. For different combinations of methods and data sets, Table S5 contains ROC curves of true positive rate and false positive rate, and tables of true positive, false positive, true negative, and false negative values for each threshold.