HeteroKGRep: Heterogeneous Knowledge Graph based Drug Repositioning.

Area of Science:

• Biomedical informatics
• Computational pharmacology
• Drug discovery

Background:

• Drug development is lengthy and expensive, with patent complexities hindering innovation.
• Existing drug repositioning models often rely on limited, homogeneous data sources.
• Heterogeneous biomedical knowledge graphs offer a richer data landscape for drug repositioning.

Purpose of the Study:

• To propose HeteroKGRep, a novel model for drug repositioning utilizing heterogeneous biomedical knowledge graphs.
• To overcome the limitations of previous models dependent on homogeneous data.
• To enhance the discovery of new therapeutic applications for existing drugs.

Main Methods:

• HeteroKGRep employs a multi-step framework involving similarity graph generation and SMOTE over-sampling.
• It utilizes a heterogeneous graph neural network to generate node sequences.
• Drug and disease embeddings are extracted for prediction of repurposing opportunities.

Main Results:

• HeteroKGRep achieved state-of-the-art performance with 99% accuracy, 95% AUC ROC, and 94% average precision.
• The model effectively leverages diverse knowledge sources for enriched representation learning.
• It demonstrates superior performance compared to existing homogeneous approaches.

Conclusions:

• HeteroKGRep establishes a promising new paradigm for knowledge-guided drug repositioning.
• The model can discover novel drug-disease associations, complementing de novo drug development.
• Utilizing multimodal biomedical data in heterogeneous graphs enhances drug repurposing strategies.