DeepMPF: deep learning framework for predicting drug-target interactions based on multi-modal representation with...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery and development

Background:

• Drug-target interaction (DTI) prediction is vital for drug design but faces challenges due to complex biological and chemical spaces.
• Traditional experimental methods for DTI prediction are time-consuming and costly.
• Existing computational methods often fail to fully leverage heterogeneous biological network structures and multi-modal information for accurate DTI prediction.

Purpose of the Study:

• To develop a multi-modal representation framework, DeepMPF, for accurate drug-target interaction prediction.
• To effectively utilize heterogeneous information and meta-path semantic analysis to improve DTI prediction accuracy.
• To provide a computational tool for prescreening potential drug candidates.

Main Methods:

• Constructed protein-drug-disease heterogeneous networks.
• Extracted feature information from sequence, heterogeneous structure, and similarity modalities.
• Employed six representative meta-path schemas for capturing high-order network structures and hidden information.
• Generated comprehensive feature descriptors and performed joint learning to predict interaction probabilities.

Main Results:

• DeepMPF achieved competitive predictive performance across four gold standard datasets.
• Drug repositioning experiments for COVID-19 and HIV demonstrated the practical applicability of DeepMPF.
• Molecular docking experiments validated the credibility of drug candidates predicted by DeepMPF.

Conclusions:

• DeepMPF effectively predicts drug-target interactions, offering a valuable tool for computational drug discovery.
• The framework enhances the efficiency of prescreening potential drug candidates for proteins.
• A publicly available web server for DeepMPF facilitates further research in the field.