MZINBVA: variational approximation for multilevel zero-inflated negative-binomial models for association analysis in...

Area of Science:

• Microbiome research
• Statistical modeling
• Bioinformatics

Background:

• The microbiome plays a critical role in human health and disease.
• Microbiome surveys face statistical challenges due to sparse, overdispersed count data and complex correlations.
• Hierarchical data structures, like those in the Human Microbiome Project (HMP), require specialized statistical approaches.

Purpose of the Study:

• To develop advanced statistical methods for association analysis in microbiome surveys.
• To address the challenges of sparse, overdispersed, and hierarchically structured microbiome data.
• To provide a robust framework for testing associations between microbial communities and environmental factors or clinical outcomes.

Main Methods:

• Proposed multilevel zero-inflated negative-binomial (MZINB) models for microbiome association analysis.
• Developed a variational approximation method for maximum likelihood estimation and inference.
• Constructed a Wald-type test statistic for robust association testing.
• Implemented the method in an R package, MZINBVA.

Main Results:

• The proposed MZINB models effectively handle sparse and overdispersed microbiome count data.
• The variational approximation method provides robust parameter estimates and covariance estimation.
• The Wald-type test statistic enables reliable association testing in complex microbiome datasets.
• Demonstrated method performance through simulations and application to the HMP dataset.

Conclusions:

• Multilevel zero-inflated negative-binomial models offer a powerful solution for microbiome association studies.
• The MZINBVA R package provides a practical tool for researchers to apply these advanced statistical methods.
• This work advances the statistical toolkit for analyzing complex microbiome data, facilitating discoveries in human health and disease.