Prediction of drug-target interactions using popular Collaborative Filtering methods

Computational approaches for predicting drug-protein interactions have gained more attention in recent years. The main reason is that a correct prediction based on screening a database of small molecules against a certain class of protein can potentially accelerate drug discovery. In this paper a popular prediction method, collaborative filtering in recommender systems, is evaluated for the prediction of drug-protein interaction. The interaction matrix for the drug-protein and the rating matrix of user-item are similar and in both cases only a small subset of the matrices are known. The CF (collaborative filtering) methods are evaluated on four classes of proteins and AUC (Area under receiver operating characteristic curve) and AUPR (Area under precision-recall curve) are reported. It is shown that collaborative filtering methods can be effective in the prediction of drug-target interaction based on the known interaction matrix. These results highlight the importance of using the known interaction matrix in order to achieve high accuracy and precision in prediction.