Improved Prediction of Drug–Target Interactions\nUsing Self-Paced Learning with Collaborative Matrix Factorization

Abstract

Identifying drug–target\ninteractions (DTIs) plays an important\nrole in the field of drug discovery, drug side-effects, and drug repositioning.\nHowever, in vivo or biochemical experimental methods for identifying\nnew DTIs are extremely expensive and time-consuming. Recently, in\nsilico or various computational methods have been developed for DTI\nprediction, such as ligand-based approaches and docking approaches,\nbut these traditional computational methods have several limitations.\nThis work utilizes the chemogenomic-based approaches for efficiently\nidentifying potential DTI candidates, namely, self-paced learning\nwith collaborative matrix factorization based on weighted low-rank\napproximation (SPLCMF) for DTI prediction, which integrates multiple\nnetworks related to drugs and targets into regularized least-squares\nand focuses on learning a low-dimensional vector representation of\nfeatures. The SPLCMF framework can select samples from easy to complex\ninto training by using soft weighting, which is inclined to more faithfully\nreflect the latent importance of samples in training. Experimental\nresults on synthetic data and five benchmark data sets show that our\nproposed SPLCMF outperforms other existing state-of-the-art approaches.\nThese results indicate that our proposed SPLCMF can provide a useful\ntool to predict unknown DTIs, which may provide new insights into\ndrug discovery, drug side-effect prediction, and repositioning existing\ndrug.