Drug-Target Binding Affinity Prediction Based on Graph Neural Networks and Word2vec

Predicting drug-target interaction (DTI) is important for drug development because drug-target interaction affects the physiological function and metabolism of the organism through bonding reactions. Binding affinity is the most important factor among many factors affecting drug-target interaction, thus predicting binding affinity is the key point of drug redirection and new drug development. This paper proposes a drug-target binding affinity (DTA) model based on graph neural networks and word2vec. In this model, the word embedding method is used to convert targets/proteins sequence into sentences containing words to capture the local chemical information of targets/proteins. Then Simplified Molecular Input Line Entry System (SMILES) is used to convert drug molecules into graphs. After feature fusion, DTA is predicted by graph convolutional networks. We conduct experiments on the Kiba and Davis datasets, and the experimental results show that the proposed method significantly improves the prediction performance of DTA.