HMT-DTI: Hierarchical meta-path learning with transformer for drug–target interaction prediction

Abstract

Drug–target interaction (DTI) prediction plays a crucial role in drug discovery and repurposing by efficiently and accurately identifying potential therapeutic targets. Existing methods face challenges in capturing high-order semantic relationships in heterogeneous graphs and effectively integrating multi-meta-path information while also suffering from low computational efficiency. To address these challenges, a pre-computation-style hierarchical meta-path learning framework named HMT-DTI is proposed. HMT-DTI can effectively capture rich semantic information about drugs and targets while ensuring high computational efficiency. Specifically, during the pre-collection stage, HMT-DTI employs a Transformer-based message passing mechanism to evaluate neighbors’ importance and adaptively collect meta-path information. The incorporation of even-relation propagation reduces redundant iterations and improves efficiency. During training, HMT-DTI adopts a hierarchical knowledge extraction strategy to evaluate the importance of multi-hop neighbors and different meta-path patterns, capturing fine-grained semantic representations of drugs and targets. HMT-DTI is evaluated on three heterogeneous biological datasets and compared with several state-of-the-art methods. The results demonstrate the superiority of HMT-DTI in DTI prediction.