Dimensionality reduction based multi-kernel framework for drug-target interaction prediction.

The prediction of novel drug-target interactions (DTIs) has intrinsic significance in drug discovery research. Wet-lab experiments of DTIs are laborious and expensive; computational methods can help minimize the complexity of identifying unknown DTIs and accelerate the drug repositioning process. Nowadays, the number of drug-target features and their interactions regularly increases, disabling traditional computational methods' prediction and analyzing ability. Therefore, developing a new robust model to derive the reduced features for effective prediction is important. Further, accurate interactions also depend on the negative drug-target pairs, and it is worthwhile to build a technique to generate perfect negative pairs. To this end, we propose a new multi-label approach, called idti-MLKdr, by introducing multi-kernel learning (MKL) based SVM for DTIs prediction with various dimensionality reduction techniques. First, we have extracted the drug-target features from chemical structures and protein sequences, applying different feature extraction methods. A new technique has been developed to construct the negative drug-target pairs based on drug-drug (or protein-protein) similarity scores. Then, three-dimensionality reduction techniques have been applied to the extracted drug-target features. Finally, we trained a multi kernel-based learner together with the reduced features and combined their prediction scores to show the final results. In this experiment, we considered auROC as an evaluation metric. The proposed method has been compared with the various existing methods under five-fold cross-validation, and the experimental results indicated that idti-MLKdr attains the best auROC for predicting DTIs. We believe that improved prediction performance will motivate the researchers for further drug development. [Display omitted] • Computational model idti-MLKdr is proposed for predicting Drug-Target Interactions. • FP2, AAC, DC and TC descriptors are used for extracting Drug-target features. • CluMS technique is proposed for resolving class imbalance problem, and three DR techniques are utilized to manage the high dimensionality. • Multi-kernel Learning (MKL) based SVM algorithm is used as a classifier. • Achieves the best prediction performance and can effectively predict the potential DTIs. [ABSTRACT FROM AUTHOR]