Learning Multi-Scale Heterogeneous Representations and Global Topology for Drug-Target Interaction Prediction

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Accurate identification of drug-target interactions (DTIs) is crucial for drug discovery and repositioning.
• Integrating heterogeneous multi-source data for DTIs presents significant challenges due to complex inter-connections and intra-similarities.

Purpose of the Study:

• To develop a robust drug-target interaction prediction model.
• To effectively integrate multi-scale attributes and global topology from heterogeneous data sources.

Main Methods:

• Construction of a drug-protein-disease heterogeneous network (RPD-Net).
• Implementation of a multi-scale pairwise deep representation learning module with novel embedding and dilation convolutions.
• Application of a global topology learning module using non-negative matrix factorization (NMF) and relational-level attention.

Main Results:

• The proposed model demonstrates superior performance compared to existing state-of-the-art methods on public datasets.
• Experimental results validate the effectiveness of the multi-scale representation and global topology learning modules.
• High top k recall rates and successful case studies highlight the method's potential in identifying drug target candidates.

Conclusions:

• The developed RPD-Net model offers an effective approach for DTI prediction by deep integration of multi-source data.
• The method shows significant promise for advancing drug discovery and drug repositioning efforts.
• The model's ability to identify potential target candidates underscores its practical utility in pharmaceutical research.