10×5-Fold Cross Validation Experiment on Drug-Target Interaction Data in DrugBank (Mar. 2014)
By using the ten trained models of DrugE-Rank in cross validation experiments, we predicted unknown drug-target interactions (for Data-1 in paper) and checked if they are in the latest DrugBank. The interactions were sorted by how many times they are predicted out of the ten models. Following 2 tables show the results obtained by target-based and drug-based cross-validation, respectively. Surprisingly, 11 out of the top 21 predicted interactions for drug-based and 14 out of the 25 predicted interactions in target-based cv can be found in the latest DrugBank. This result also indicates the high predictive performance of DrugE-Rank.