ELDTIP: An Ensemble Learning-based method for DTI Prediction

Abstract

Exploring drug-target interactions has always been an important step in drug development. However, exploring drug-target interaction is time-consuming and laborious. A large number of studies try to use artificial intelligence methods to predict possible drug-target interactions to reduce the workload of the wet-lab identification experiments. However, the accuracy of existing methods is still limited. This paper proposes an ensemble learning-based drug-target interaction prediction method (ELDTIP in short). First, the multiple similarity matrices of drugs or proteins are integrated by singular value decomposition (SVD) to obtain their low-dimensional feature vectors. After that, by concatenating the low-dimensional feature vectors of specific drugs and targets, the feature vector of a drug-target pair are obtained. An ensemble learning model based on gradient boosting decision tree (GBDT) was constructed to predict whether this pair of drug-target can interact with each other. The main contributions of ELDTIP are as follows: (1): ELDTIP uses SVD to integrate multiple similarity matrices, which can retain more valuable information of the original feature. (2): ELDTIP uses the ensemble learning-based model, GBDT, which can make full use of the unknown DTIs in the dataset and mitigate the influence of class imbalance. Experimental results show that the performance of ELDTIP is higher than that of several state-of-the-art DTI prediction methods.