A Novel Drug-Target Interaction Predication Model Via Singular Value Decomposition and Collaborative Filtering

Since drug repositioning can reduce the time for drug development and accelerate the drugs into clinical phases, it has become a hot topic in computational biology. Identifying potential drug-target interactions (DTIs) is a basic and important task in drug repositioning. Many methods have been proposed to predict DTIs, while there still has a large room to improve the prediction accuracy. In this paper, a novel dual network matrix factorization model based on singular value decomposition and collaborative filtering is proposed to predict DTIs. First of all, multiple kernels are built from different information sources. Then, singular value decomposition is applied to the DTIs matrix to get the singular value matrix and drug-specific and target-specific latent vectors. Finally, a collaborative filtering method is employed to identify potential DTIs. The results show that the proposed model outperforms other state-of-the-art in silico methods on four benchmark datasets in terms of the area under precision and recall (AUPR) scores.