Predicting Drug-target Interaction via Wide and Deep Learning

Abstract

Identifying the interactions of approval drugs and targets is essential in medicine field, which can facilitate the discovery and reposition of drugs. Due to the tendency towards machine learning, a growing number of computational methods have been applied to the prediction of the drug-target interactions (DTIs). In this paper, we propose a wide and deep learning framework combining a generalized linear model and a deep feed-forward neural network to address the challenge of predicting the DTIs precisely. The proposed method is a joint training of the wide and deep models, which is implemented by feeding the weighted sum of the results obtained from the wide and deep models into a logistic loss function using mini-batch stochastic gradient descent. The results of this experiment indicate that the proposed method increases the accuracy of prediction for DTIs, which is superior to other methods.