Drug discovery is the process through which every drug that is consumed today has been discovered. It takes around a decade for a new drug to be discovered, with the average cost that is estimated at $2.5 billion. As a result, drug developers attempt to find new uses for existing and/or released drugs to facilitate the process of drug development and diminish associated costs which is called drug repurposing.
Drugs have specific effects on proteins in the human's body called targets. Predicting and finding new targets for existing drugs can result in drug repurposing. The interactions between drugs and targets can be represented by bipartite graphs with two set of nodes, drugs and targets, where edges represent the interactions. However, conventional approaches fail to efficiently analyze the drug-target interaction datasets which has resulted in application of machine-learning based methods in all drug development stages, including the design of a new drug, determining new usage for current drug, identifying drug targets in the body and more.
In recent years, several studies have been carried out on graph neural networks, resulting in remarkable progress in the function of tasks with data having the graph structure, which is necessary for recommender applications. Such applications extract features of the network by building a model, and then predict the existence and absence of a link between them. A central factor in computational drug discovery is to predict whether a specified drug can attach a distinct protein or not. Here, we attempt to examine the problem of the existence or absence of a link in drug-target interactions, using neural networks, graph illustration, and previous approaches in recommender systems. The results of our study may be a useful tool for further developments in the drug industry.