Explore potential disease related metabolites based on latent factor model

Area of Science:

• Metabolomics
• Computational Biology
• Biomedical Informatics

Background:

• Metabolomics aids in discovering disease signatures and elucidating molecular mechanisms.
• Identifying disease-related metabolites is crucial for understanding pathogenesis and advancing clinical medicine.

Purpose of the Study:

• To develop a computational model for predicting potential associations between metabolites and diseases.
• To identify novel metabolic signatures linked to human diseases.

Main Methods:

• A disease and literature driven metabolism prediction model (DLMPM) was developed using a latent factor model.
• An association matrix was constructed using disease terms from various databases and metabolite mappings, completed using similarity measures.
• Matrix decomposition was employed to predict metabolite-disease associations.

Main Results:

• The DLMPM identified 1,406 direct associations and predicted 119,206 unknown associations between metabolites and diseases with 80.88% coverage.
• Model performance was validated on 19 diseases, achieving an average Area Under the Curve (AUC) of 82.33%.

Conclusions:

• The proposed DLMPM effectively predicts potential metabolite-disease pairs.
• The model demonstrates superior performance in prioritizing candidate disease-related metabolites compared to existing methods, offering valuable insights for researchers.