BMCMDA: a novel model for predicting human microbe-disease associations via binary matrix completion.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• The human microbiome significantly influences health, with many noninfectious diseases linked to microbial communities.
• Identifying microbe-noninfectious disease associations (MDAs) is challenging due to high costs and cultivation limitations.
• Computational methods, particularly machine learning, offer a promising avenue for rapid MDA prediction.

Purpose of the Study:

• To develop a novel computational approach for predicting potential microbe-noninfectious disease associations (MDAs).
• To address the limitations of traditional methods in identifying MDAs by leveraging machine learning.

Main Methods:

• Proposed a novel predictive approach named Binary Matrix Completion for microbe-noninfectious disease associations (BMCMDA).
• BMCMDA models the incomplete microbe-disease association matrix using latent parameterizing and noising matrices.
• Employed probit regression to model relationships and a binomial model for observed entries.

Main Results:

• BMCMDA achieved high performance in leave-one-out cross-validation with an AUC of 0.906 and AUPR of 0.526.
• Demonstrated superior prediction accuracy, with approximately 7% and 5% improvements in AUC and AUPR, respectively, over the KATZHMDA approach.
• Effectively predicts the likelihood of a microbe being associated with a specific disease.

Conclusions:

• BMCMDA offers an effective and efficient method for predicting microbe-noninfectious disease associations.
• The approach can be extended to other binary relationship prediction tasks, such as protein-protein interactions and drug-target interactions.