MPTN: A message-passing transformer network for drug repurposing from knowledge graph.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Drug Discovery

Background:

• Drug repurposing (DR) leverages existing drugs for new therapeutic uses, often guided by knowledge graphs (KGs).
• Current KG-based DR methods struggle with extracting rich semantic information from complex biomedical data.
• Advancements in computing and data availability necessitate improved KG reasoning models for DR.

Purpose of the Study:

• To propose a novel message-passing transformer network (MPTN) for enhanced drug repurposing.
• To improve the extraction of semantic information from contextual triples in biomedical knowledge graphs.
• To predict new therapeutic pathways for existing drugs more accurately.

Main Methods:

• Developed a message-passing transformer network (MPTN) integrating CompGCN for embedding aggregation and a transformer-based message passing module.
• Incorporated attention mechanisms to capture semantic context within entity triples.
• Utilized residual connections for information preservation and InteractE for heterogeneous feature interaction prediction.

Main Results:

• The proposed MPTN model demonstrated superior performance compared to existing knowledge graph embedding (KGE) methods.
• Experiments on two datasets confirmed the model's effectiveness in predicting new drug indications.
• The MPTN model showed significant improvements in extracting semantic information from biomedical KGs.

Conclusions:

• MPTN offers a powerful new approach for knowledge graph-based drug repurposing.
• The model's ability to capture contextual semantic information enhances the prediction of novel therapeutic applications for drugs.
• This work advances the field of computational drug discovery by improving KG reasoning capabilities.