CMFHMDA: a prediction framework for human disease-microbe associations based on cross-domain matrix factorization.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Identifying microbe-disease associations is crucial for developing new health strategies.
• Laboratory verification of these associations is costly and time-consuming.
• There is a need for efficient computational tools to predict microbe-disease links.

Purpose of the Study:

• To develop a novel computational algorithm, Cross-Domain Matrix Factorization for Human Disease-Microbe Associations (CMFHMDA), for predicting microbe-disease associations.
• To leverage network data and matrix factorization for enhanced prediction accuracy.

Main Methods:

• Calculated composite disease similarity and Gaussian interaction profile similarity for microbes.
• Utilized the Weighted K Nearest Known Neighbors (WKNKN) algorithm to refine the microbe-disease association matrix.
• Integrated cross-domain network data of microbes and diseases within the CMFHMDA model.

Main Results:

• The CMFHMDA model demonstrated high accuracy across multiple cross-validation methods.
• Achieved Area Under the ROC Curve scores of 0.9172 (global LOOCV), 0.8551 (local LOOCV), and 0.9351 (5-fold CV).
• Many predicted microbe-disease associations were validated by existing experimental studies.

Conclusions:

• CMFHMDA is an effective computational framework for predicting microbe-disease associations.
• The model's accuracy and validation suggest its utility in identifying potential disease-associated microbes.
• This approach can accelerate the discovery of microbial targets for diagnostics and therapeutics.