A Framework f or Predicting Drug Target Interaction Pairs Through Heterogeneous Information Fusion

Abstract

Drugs, also known as medicines cure diseases by interacting with some specific targets such as proteins and nucleic acid. Prediction of such drug-target interaction pairs plays a major role in drug discovery. It helps to identify the side effects caused by various drugs and provide a way to analyze the chances of usage of one drug for various diseases apart from the one disease that is predefined for that drug. However, existing Drug Target Interaction prediction methods are very expensive and time consuming. In this work, we present a new method to predict such interactions with the help of bipartite graph, which represents the known drug target interaction pairs. Information about drug and target are collected from various sources and they are integrated using Kronecker Regularized Least Square approach and Multiple Kernel Learning method, to generate drug and target similarity matrices. By integrating the two similarity matrices and known DTIs a heterogeneous network is constructed and new DTI predictions are done by performing Bi Random walk in it.