Bayesian variable selection for multivariate zero-inflated models: Application to microbiome count data.

Area of Science:

• Microbiology
• Statistical Modeling
• Bioinformatics

Background:

• Microorganisms critically influence human health and disease.
• Interactions within microbial communities are complex, necessitating multivariate statistical approaches.
• Microbial count data often exhibit zero-inflation, posing analytical challenges.

Purpose of the Study:

• To develop a Bayesian variable selection model for multivariate count data with excess zeros.
• To incorporate covariance structure and estimate associations with mean levels of microbial taxa.
• To address limitations in existing models for high-dimensional, multivariate, zero-inflated data.

Main Methods:

• Developed a Bayesian variable selection model for multivariate count data.
• Incorporated information on the covariance structure of multiple taxa counts.
• Compared the proposed method against univariate approaches using simulations.

Main Results:

• The proposed multivariate method maintained Type I error and improved detection of true associations in the binary component when outcomes were correlated.
• Univariate methods sometimes showed higher power for the count component but at the cost of inflated false discovery rates.
• The approach successfully identified five oral microbial species associated with HIV infection in a real-world study.

Conclusions:

• The developed Bayesian model effectively handles zero-inflated, multivariate microbial count data.
• This approach offers a robust alternative to univariate analyses, particularly for correlated microbial data.
• The method has practical applications in identifying microbial biomarkers for diseases like HIV.