Drug-Target Affinity Prediction Based on Multi-channel Graph Convolution

Computer-aided drug design with high performance is a promising field, and the pre-diction of drug target affinity is an important part in computer-aided drug design. As a kind of deep learning model algorithm, graph neural network series model algorithm has been gradually applied in drug and protein research due to its excellent performance in structural feature learning. In the new field of drug target affinity prediction, graph neural network also has great potential. In this paper, a novel approach for drug target affinity prediction based on multi-channel graph convolution network is proposed. The method encodes drug and protein sequences into corresponding node adjacency matrix. The adjacency matrix together with the physical and chemical characteristics of drug and protein sequences are used as the inputs of the model to construct a multi-channel graph convolution network that aggregates the information of nodes at different distances. The drug feature and target feature vectors are concatenated, and then through the full connection layer, the concatenated vector is converted to the predicted value. The experiment results on Davis dataset and KIBA dataset show that the proposed method outperforms most relevant methods. While the experimental results is Slightly worse than GraphDTA, it shows that the proposed method can improve the prediction of drug-target affinity to a certain extent and aggregate more information from other nodes of higher order proximity in the graphs.