Infants' gut microbiome data: A Bayesian Marginal Zero-inflated Negative Binomial regression model for multivariate...

Area of Science:

• Microbiome research
• Statistical modeling
• Infant health

Background:

• Infant gut microbiome composition exhibits high inter-individual variability.
• Existing statistical methods may not fully capture the dynamic nature and complexities of infant microbiome data.
• Next-generation sequencing technologies require advanced analytical approaches.

Purpose of the Study:

• To propose and evaluate a novel statistical model, Bayesian Marginal Zero-inflated Negative Binomial (BAMZINB), for analyzing infant gut microbiome data.
• To address challenges of zero-inflation, over-dispersion, and multivariate structures inherent in microbiome data.
• To compare the performance of BAMZINB against existing methods using simulations and real-world data.

Main Methods:

• Developed the Bayesian Marginal Zero-inflated Negative Binomial (BAMZINB) model.
• Conducted 32 simulation scenarios to compare BAMZINB with glmFit and BhGLM.
• Applied the BAMZINB model to the SKOT cohort (I and II) real-world infant microbiome dataset.

Main Results:

• BAMZINB demonstrated comparable performance to glmFit and BhGLM in estimating average abundance differences.
• BAMZINB showed a better fit in most simulation scenarios, particularly with large sample sizes and strong signals.
• Analysis of the SKOT cohort revealed significant changes in bacterial abundance between 9 and 18 months in infants of healthy versus obese mothers.

Conclusions:

• The BAMZINB model is recommended for analyzing infant gut microbiome data due to its ability to handle zero-inflation and over-dispersion.
• This approach provides a robust framework for multivariate analysis and comparison of average bacterial abundance differences.
• Findings highlight dynamic microbiome shifts in infants related to maternal health status.