Multi-network logistic matrix factorization for metabolite-disease interaction prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Systems biology

Background:

• Identifying metabolite-disease interactions is crucial for disease diagnosis, prevention, and treatment.
• Existing methods struggle with novel diseases and metabolites due to limited known associations.
• Integrating diverse omics data can enhance the prediction of these complex relationships.

Purpose of the Study:

• To propose a novel computational model, multiple-network logistic matrix factorization (MN-LMF), for predicting metabolite-disease interactions.
• To improve the prediction accuracy for both known and unknown metabolite-disease associations.
• To address the challenge of identifying interactions for new diseases and metabolites.

Main Methods:

• Constructing disease and metabolite similarity networks by integrating heterogeneous omics data.
• Employing modified logistic matrix factorization to combine similarity networks with known interaction data.
• Developing a computational framework for predicting potential metabolite-disease interactions.

Main Results:

• MN-LMF demonstrates high accuracy in predicting metabolite-disease interactions.
• The proposed model outperforms existing state-of-the-art methods in prediction performance.
• Case studies confirm the model's effectiveness in inferring unknown interactions for novel diseases.

Conclusions:

• MN-LMF is an effective computational tool for predicting metabolite-disease interactions.
• The model shows significant promise for advancing disease diagnosis and therapeutic strategies.
• This approach facilitates the discovery of novel metabolite-disease links, particularly in understudied areas.