Collaborative Matrix Factorization For Predicting Drug-Target Interactions

Research about the interactions between drugs and targets is very important for the research and development of drugs. Traditional chemical experiments have low efficiency and high cost. On the other hand, research by computer technology has the advantage of high efficiency and low money expenses. The latter has already become the main research approach in this field. This article focuses on machine learning-based approaches, particularly similarity-based methods for drug-target interaction prediction. In this field, the development of new prediction methods with high accuracy is the hotspot.In this article, we first review several machine learning-based methods for predicting drug-target interactions that are currently state-of-the-art and have shown good predicting performances. We then propose a new machine learning method based on matrix factorization, called Collaborative Matrix Factorization (CMF), and compare this method with the previously reviewed methods under uniform experimental settings. As a result, CMF preforms the best in 9 out of 12 settings, showing its outstanding advantage in drug-target interaction prediction. Finally, we verify the newly predicted drug-target pairs by CMF in several latest databases. Results show that 14 out of 20 new predictions are verified, showing the practical value of CMF.