Predicting drug-drug interactions based on integrated similarity and semi-supervised learning

Abstract

A drug-drug interaction (DDI) is 
defined as an association between two drugs 
where the pharmacological effects of a drug 
are influenced by another drug. Positive 
DDIs can usually improve the therapeutic 
effects of patients, but negative DDIs cause 
the major cause of adverse drug reactions 
and even result in the drug withdrawal from 
the market and the patient death. Therefore, 
identifying DDIs has become a key 
component of the drug development and 
disease treatment. In this study, we propose
a novel method to predict DDIs based on the 
integrated similarity and semi-supervised 
learning (DDI-IS-SL). DDI-IS-SL integrates 
the drug chemical, biological and phenotype 
data to calculate the feature similarity of 
drugs with the cosine similarity method.

The Gaussian Interaction Profile 
kernel similarity of drugs is also calculated
learning method (the Regularized Least 
Squares classifier) is used to calculate the 
interaction possibility scores of drug-drug 
pairs. In terms of the 5-fold cross validation, 
10-fold cross validation and de novo drug 
validation, DDI-IS-SL can achieve the better 
prediction performance than other 
comparative methods. In addition, the 
average computation time of DDI-IS-SL is 
shorter than that of other comparative 
methods. Finally, case studies further 
demonstrate the performance of DDI-IS-SL 
in practical applications.