HS-DTI: Drug-target Interaction Prediction Based on Hierarchical Networks and Multi-order Sequence Effect

Abstract

Computer-aided drug design utilizes powerful computers to simulate the tasks in drug design, which could significantly accelerate drug development and reduce cost consumption. Drug-target interaction (DTI) prediction is the critical stage in computer-aided drug design and has drawn increased interest from both academia and pharmacogenomics. A variety of computational models for DTI have been available that rely only on the linear representation of compounds and amino acid sequences. However, the significant drawbacks are that these models do not consider the structures and functions of specific functional groups in drugs as well as the multi-order sequence effect of proteins. In this study, we present a novel computational model named HS-DTI that attempts to extract plentiful features of drugs and proteins by capturing functional group structures and incorporating multi-order sequence information, respectively. Specifically, the probability assignment matrix is learned through the hierarchical network module to identify the functional group structures in reconstructed drug graphs on the drug encoder. Next, modelling 1st-order and 2nd-order sequence information based on the hydrophilicity, hydrophobicity, and side-chain mass of the most contiguous and the second-most contiguous amino acid residues. Finally, we comprehensively evaluated the performance of HS-DTI with several state-of-the-art methods on real-world datasets. Experimental results clearly show the effectiveness of the proposed model. Additionally, ablation experiments are also being devised to validate this performance improvement.