Adaptive Graph Prompting Meets Contrastive Learning: A Multi-View Framework for Metabolite-Disease Association...

Area of Science:

• Computational biology
• Bioinformatics
• Precision medicine

Background:

• Metabolite-disease associations (MDAs) are crucial for precision medicine.
• Existing computational methods struggle with data sparsity, noise, and feature representation.

Purpose of the Study:

• To propose GPLCL (graph prompt-enhanced contrastive learning), a novel multi-view graph learning framework.
• To address challenges in MDA identification using adaptive graph prompting and contrastive learning.

Main Methods:

• GPLCL integrates enhanced graph prompt features (GPF+) with attention-based node adaptation.
• Utilizes strategic graph augmentation and self-supervised contrastive optimization (HeteroGraphSAGE) for topological invariant preservation and multi-scale pattern aggregation.

Main Results:

• Achieved AUC 0.9761 and AUPR 0.9729 on Dataset 1, outperforming existing methods by 0.55–6.37%.
• Maintained strong performance (AUC 0.9576, AUPR 0.9499) on noisy Dataset 2, demonstrating robustness.
• Case studies showed potential in discovering novel MDAs for type 1 diabetes, obesity, and Parkinson's disease.

Conclusions:

• GPLCL offers a robust and effective framework for identifying metabolite-disease associations.
• The model shows significant potential for advancing metabolomics research and translational medicine.
• The developed computational framework enhances precision medicine by improving MDA prediction accuracy and robustness.