Multi-View Feature Aggregation for Predicting Microbe-Disease Association.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Microbes significantly impact human health and disease, necessitating accurate identification of disease-related microbes for therapeutic applications.
• Developing robust computational methods is crucial for understanding microbe-disease relationships.
• Existing methods may not fully capture the complex interplay between microbes and diseases.

Purpose of the Study:

• To propose and validate a novel computational scheme, Multi-View Feature Aggregation (MVFA), for identifying disease-related microbes.
• To integrate diverse features, including linear and nonlinear representations, for enhanced prediction accuracy.
• To improve the prediction of human microbe-disease associations.

Main Methods:

• Developed a Multi-View Feature Aggregation (MVFA) scheme integrating linear and nonlinear features.
• Employed non-negative matrix tri-factorization (NMTF) and a bi-random walk model for linear feature extraction.
• Utilized a capsule neural network for nonlinear feature extraction and logistic regression for final score aggregation.

Main Results:

• The MVFA scheme demonstrated superior performance in predicting microbe-disease associations compared to state-of-the-art methods on two datasets.
• Experimental results showed improved recovery of missing associations and prediction for novel microbes.
• Ablation studies confirmed the benefit of aggregating multi-view linear and nonlinear features.

Conclusions:

• The proposed MVFA scheme effectively identifies disease-related microbes by integrating multi-view features.
• This computational approach offers a promising tool for advancing microbe-disease association research and potential therapeutic strategies.
• Case studies on Type 1 diabetes and Liver cirrhosis validated the method's practical effectiveness.