KNDM: A Knowledge Graph Transformer and Node Category Sensitive Contrastive Learning Model for Drug and Microbe...

Area of Science:

• Computational biology and bioinformatics
• Pharmacology and drug discovery
• Microbiome research

Background:

• The human microbiome significantly influences drug efficacy and toxicity.
• Accurate identification of drug-related microbes is crucial for understanding drug mechanisms.
• Existing graph learning methods for drug-microbe prediction have limitations in utilizing diverse entity characteristics and meta-path contextual relationships.

Purpose of the Study:

• To propose a novel model, KNDM, that addresses limitations in current drug-microbe association prediction.
• To enhance the utilization of knowledge graph features and meta-path contextual relationships for improved prediction accuracy.
• To develop a method that ensures consistency between entity features and node semantic features.

Main Methods:

• Construction of a comprehensive knowledge graph integrating drug and microbe entities.
• Development of an entity category-sensitive transformer to handle entity heterogeneity and relationships.
• Implementation of a meta-path semantic feature learning strategy with recursive gating.
• Application of a node-category-sensitive contrastive learning strategy to improve feature consistency.

Main Results:

• The proposed KNDM model significantly outperforms eight state-of-the-art drug-microbe association prediction models.
• Ablation studies confirm the effectiveness of KNDM's core innovations, including the transformer and contrastive learning components.
• Case studies demonstrate KNDM's ability to identify potential associations for drugs like curcumin, epigallocatechin gallate, and ciprofloxacin.

Conclusions:

• KNDM provides a robust framework for predicting drug-microbe associations by effectively leveraging knowledge graphs and advanced learning techniques.
• The model's ability to integrate diverse features and contextual information leads to superior predictive performance.
• KNDM offers a valuable tool for advancing research in personalized medicine and understanding drug-host-microbe interactions.