An Efficient Computational Method to Predict Drug-Target Interactions Utilizing Matrix Completion and Linear Optimization Method

The experimental determination of drug-target interaction is time-consuming and expensive. Therefore, a continuous demand for more effective prediction of drug-target interaction using computing technology. Many algorithms have been designed to infer potential interactions. Most of these algorithms rely on drug similarity and target similarity as auxiliary information in modeling, but they ignore the problem that there are a lot of missing auxiliary data of existing drugs or targets, which affects the prediction performance of the model and fails to achieve the expected effect. Here, we propose a calculation model named MCLO, which is based on the matrix completion and linear optimization technology to predict novel drug-target interactions. First, the proposed method calculate the side effect similarity of drugs and protein-protein interaction similarity of targets. Then we utilize the idea of linear neighbor representation learning to predict the potential drug-target interaction. It is worth mentioning that our method uses the idea of matrix completion technology to complete the imperfect similarity before drug-target interactions prediction. To evaluate the performance of MCLO, we carry out experiments on four gold standard datasets. The experimental results show that MCLO can be effectively applied to identify drug-target interactions.