Progressive Orthogonal Multimodal Similarity Learning for Metabolite-Disease Association Prediction.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Systems Biology

Background:

• Identifying metabolite-disease associations (MDAs) is vital for understanding disease mechanisms.
• Existing methods often fail to fully exploit complementary information or higher-order relationships between metabolites and diseases.

Purpose of the Study:

• To develop an advanced computational method for predicting metabolite-disease associations (MDAs).
• To address limitations in current methods by effectively mining multimodal similarity and higher-order relationships.

Main Methods:

• Proposed Progressive Orthogonal Multimodal Similarity Learning (POMSL) method.
• Constructed hypergraphs using KNN and K-means to capture higher-order relationships.
• Applied hierarchical contrastive learning for multi-similarity feature enhancement.
• Utilized a progressive orthogonal multimodal similarity integration strategy and a bilinear decoder for MDA prediction.

Main Results:

• POMSL demonstrated excellent performance in metabolite-disease association prediction.
• Validated the effectiveness of the proposed hierarchical contrastive learning approach.
• Confirmed the benefits of the progressive orthogonal multimodal similarity integration strategy.

Conclusions:

• POMSL offers a powerful new approach for predicting metabolite-disease associations.
• The method effectively integrates multimodal similarity information and captures complex biological relationships.
• The findings highlight the potential of advanced machine learning techniques in advancing our understanding of disease etiology.