Generalized matrix factorization based on weighted hypergraph learning for microbe-drug association prediction.

Area of Science:

• Microbiology
• Pharmacology
• Computational Biology

Background:

• Microbial communities significantly impact human health, drug development, and precision medicine.
• Identifying microbe-drug associations is crucial for drug discovery, therapy, and understanding microbial mechanisms.
• Computational methods offer an efficient alternative to costly biological experiments for predicting these associations.

Purpose of the Study:

• To develop a computational model for predicting potential microbe-drug associations.
• To leverage multi-omics data and hypergraph learning for enhanced prediction accuracy.
• To provide a tool that complements experimental methods in microbe-drug association research.

Main Methods:

• Proposed a generalized matrix factorization model based on weighted hypergraph learning (WHGMF).
• Integrated multi-omics data to compute microbe and drug features, including functional, semantic, and structural similarities.
• Constructed a hypergraph using neighborhood information and calculated hyperedge weights using simple volume.
• Incorporated hypergraph regularization and high-order structural information into the matrix factorization model.

Main Results:

• WHGMF accurately predicts potential microbe-drug associations.
• The model demonstrates adaptability to class-imbalanced datasets.
• WHGMF is effective for predicting associations involving novel microbes and drugs.
• A case study validated the method's effectiveness.

Conclusions:

• WHGMF provides an accurate and efficient computational approach for predicting microbe-drug associations.
• The method enhances drug discovery and precision medicine by identifying key microbial interactions.
• WHGMF offers a valuable tool for researchers exploring the complex relationship between microbes and drugs.