MHANDTI: Drug-Target Interaction Prediction Model Based on Heterogeneous Graph Multi-Hop Attention Networks

Predicting drug-target interactions (DTI) has become an important step in the drug discovery and drug repositioning process. The biological identification of DTI incurs significant financial and temporal costs, and the deployment of computational methodologies for DTI prediction can substantially curtail both the duration and economic expenditure of drug discovery or repositioning. Drawing from disparate data sources can offer a comprehensive overview and diverse perspectives for predicting drug-target interactions. However, existing methods for processing drug and target information are unable to capture long-range dependency across heterogeneous graphs. To capture more comprehensive drug and target features for DTI prediction, this article proposes a drug-target interaction prediction model based on heterogeneous graph multi-hop attention networks, called MHANDTI. Concretely, the word-level deep convolutional neural network is used to obtain the structural features of drugs and targets as attributes of drug and target entity nodes in heterogeneous graphs. For the processing of heterogeneous graphs, MHANDTI designs a multi-hop attention diffusion layer (MHADL) to establish a connection between the non-direct connection nodes, and summarize the characteristics of the adjacent nodes of the center nodes in several hops. A series of results indicate that the performance of this method is superior to the latest existing frameworks.