IDNDDI: An Integrated Drug Similarity Network Method for Predicting Drug-Drug Interactions

Abstract

A drug-drug interaction(DDI) was defined as the pharmacological effect(s) of a drug influenced by another drug. The positive DDIs can improve the therapeutic effect of patients. However, the negative DDIs can lead serious results, such as drug withdrawal from market and even patient death. Currently, multiple pharmaceutical drugs have widely been used to treat complex diseases, such as cancer. The traditional biomedical experiments are very time-consuming and very costly to validate new DDIs. Therefore, it is appealing to develop computational methods to discover potential DDIs. In this study, we propose a new computational method (called IDNDDI) to predict novel DDIs. Based on the binary vector of drug chemical, biological and phenotype data, IDNDDI computes the integrated drug feature similarity by the cosine similarity method. In addition, the node-based drug network diffusion method is used to calculate the relational initial scores for new drugs. To systematically evaluate the prediction performance of IDNDDI and compare it with other prediction methods, we conduct the 5-fold cross validation and de novo drug validation. In terms of the AUC (area under the ROC curve)value, IDNDDI achieves the better prediction performance in the 5-fold cross validation, specifically, the AUC value is 0.9691, which is larger than the state-of-the-art L1E (L1 Classifier ensemble method) results of 0.9570. In addition, IDNDDI also obtains the best prediction result in the de novo drug validation and the AUC reaches 0.9292. The prediction ability in application of our method is also illustrated by case studies. IDNDDI is an effective DDI prediction method which can help to reduce adverse drug reactions and improve the efficiency of drug development progress.