Inferring drug-target interactions based on random walk and convolutional neural network.

Computational strategies for identifying drugtarget interactions (DTIs) can guide the process of drug discovery. Most proposed methods predict DTIs via integration of heterogeneous data related to drugs and proteins. However, they have failed to deeply integrate these data and learn deep feature representations of multiple original similarities and interactions. We constructed a heterogeneous network by integrating various connection relationships, including drugs, proteins, and drug side effects and their similarities, interactions, and associations. A prediction method, DTIPred, was proposed based on random walk and convolutional neural network. DTIPred utilizes original features related to drugs and proteins and integrates the topological information. The random walk is applied to construct the topological vectors of drug and protein nodes. The topological representation is learned by the learning frame based on convolutional neural network. The model also focuses on integrating multiple original similarities and interactions to learn the original representation of the drugprotein pair. The experimental results demonstrate DTIPred has better prediction performance than several state-of-the-art methods. It can retrieve more actual drugprotein interactions in the top part of the predicted results, which may be more helpful to biologists. Case studies on five drugs demonstrated DTIPred could discover potential drugprotein interactions.