Graph convolution network based on meta-paths and mutual information for drug-target interaction prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Network Science

Background:

• Drug-target interactions (DTIs) are crucial for drug repositioning, but prediction faces challenges like data sparsity.
• Deep learning methods struggle with inadequate representation of feature relationships in complex networks.

Purpose of the Study:

• To develop a novel graph convolutional network, GCNMM, for predicting latent DTIs.
• To address sparsity and feature representation issues in drug-target heterogeneous networks.

Main Methods:

• Constructed a fused DTI network using meta-paths and a graph attention network.
• Integrated similarity networks via entropy-based fusion and employed a graph convolutional auto-encoder.
• Incorporated spatial topological consistency and mutual information maximization for dual optimization.

Main Results:

• GCNMM demonstrated superior performance compared to existing baseline models in DTI prediction.
• Experimental results confirmed the effectiveness of the proposed GCNMM model.

Conclusions:

• GCNMM shows significant potential for accurate DTI prediction.
• The model's practical effectiveness was validated through case studies, supporting its application in drug repositioning.