NBZIMM

NBZIMM implements zero-inflated negative binomial mixed modeling to analyze longitudinal metagenomic count data, addressing high-dimensionality, sample dependence, overdispersion, and excess zeros for inference on microbial community dynamics.

Key Features:

• Zero-Inflated Negative Binomial Mixed Modeling (ZINBMM): Employs zero-inflated negative binomial mixed models to handle high-dimensional metagenomic count data with overdispersion, excess zeros, and sample dependence.
• Fast EM-IWLS Algorithm: Fits ZINBMMs using a fast Expectation-Maximization Iteratively Weighted Least Squares (EM-IWLS) algorithm to improve computational efficiency relative to numerical integration methods.
• Flexibility in Modeling: Supports inclusion of fixed and random effects and specification of within-subject correlation structures for longitudinal analyses.
• Superior Performance: Demonstrates improved computational efficiency and statistical accuracy versus linear mixed models, negative binomial mixed models, and zero-inflated Gaussian mixed models in simulations and real data.

Scientific Applications:

• Longitudinal human microbiome studies: Enables analysis of temporal changes in microbial community composition using metagenomic count data.
• Disease mechanism and biomarker discovery: Facilitates investigation of microbiome associations with health and disease and supports identification of potential biomarkers.

Methodology:

Fits zero-inflated negative binomial mixed models using a fast Expectation-Maximization Iteratively Weighted Least Squares (EM-IWLS) algorithm to accommodate overdispersion, excess zeros, mixed fixed and random effects, and within-subject correlation.