DVMMHGNN: A Dual View Multi-Modal Heterogeneous Graph Neural Network with Contrastive Learning for Microbe-Informed...

Area of Science:

• Computational Biology
• Pharmacology
• Bioinformatics

Background:

• Drug repurposing (DR) accelerates pharmaceutical development by finding new uses for existing drugs.
• Accurately identifying drug-disease associations is crucial but challenging due to complex biological interactions.
• Current methods often neglect the microbiota's regulatory role and struggle with semantic consistency in data integration.

Purpose of the Study:

• To propose DVMMHGNN, a microbe-informed heterogeneous graph contrastive learning framework for enhanced drug repurposing.
• To address limitations in existing computational approaches, particularly regarding microbiota influence and semantic consistency.

Main Methods:

• Developed DVMMHGNN, integrating structural and meta-path information using a heterogeneous graph contrastive learning framework.
• Employed a multimodal feature fusion module for cross-modal entity embedding.
• Utilized a graph-masked autoencoder for high-order representation learning.
• Applied contrastive learning at structural and meta-path levels to enhance semantic coherence.

Main Results:

• DVMMHGNN significantly outperformed nine state-of-the-art methods in predicting drug-disease associations.
• Achieved superior performance metrics, including AUC, AUPR, and F1-score.
• Ablation studies confirmed the effectiveness of individual model components.

Conclusions:

• DVMMHGNN offers a powerful approach for microbe-informed drug repurposing, improving prediction accuracy.
• The framework effectively captures complex biological semantics for better drug-disease association identification.
• DVMMHGNN has the potential to discover novel drug indications and guide therapeutic strategies.