Testing latent classes in gut microbiome data using generalized Poisson regression models.

Area of Science:

• Microbiome Research
• Statistical Modeling
• Human Health

Background:

• Human microbiome research is crucial for understanding health and disease.
• Microbiome data (e.g., operational taxonomic unit counts) often exhibit over-dispersion and zero-inflation.
• Existing models like generalized Poisson and zero-inflated generalized Poisson models address these data challenges.

Purpose of the Study:

• To introduce a novel statistical testing methodology for detecting latent classes in generalized Poisson regression models.
• To address the issue of zero-inflation arising from population heterogeneity in microbiome studies.
• To provide a robust method for analyzing complex human microbiome data.

Main Methods:

• Development of a closed-form test statistic for latent class detection.
• Deduction of the asymptotic distribution of the test statistic using estimating equations.
• Extensive simulation studies to evaluate the methodology's performance.
• Application of the test to real-world human gut microbiome data from the Bogalusa Heart Study.

Main Results:

• The novel testing methodology effectively identifies latent classes in generalized Poisson regression models.
• Simulation studies demonstrate the efficacy and robustness of the proposed test statistic.
• The method successfully detected latent classes within the human gut microbiome data.

Conclusions:

• The developed statistical test offers a powerful tool for uncovering hidden structures in microbiome data.
• This approach enhances the analysis of zero-inflated and over-dispersed microbiome datasets.
• The findings contribute to a better understanding of human health and disease through microbiome research.