Drug-Drug Interactions Prediction Using Fingerprint Only

Figure 2

Whole construction procedures of the classifier for predicting drug-drug interactions (DDIs). Validated DDIs are retrieved from DrugBank, which are termed positive samples. Drugs involved in positive samples are used to randomly generate negative samples. Drugs are represented by fingerprint features, which are further refined as DDI features using three models. Random forest is adopted to build the classifier, which is evaluated by two types of tenfold cross-validation.