A Bayesian longitudinal trend analysis of count data with Gaussian processes.

Area of Science:

• Statistics
• Bioinformatics
• Computational Biology

Background:

• Analyzing longitudinal count data with high variability and zero inflation poses challenges for traditional methods.
• Difficulties in characterizing time trends in datasets such as gene expression or bacteria counts are common.
• Existing methods like generalized estimating equations may struggle with these data complexities.

Purpose of the Study:

• To propose a robust Bayesian methodology for analyzing highly variable, zero-inflated longitudinal count data.
• To effectively model complex time trends and test for group differences in such datasets.
• To provide a flexible framework applicable to next-generation sequencing (NGS) data.

Main Methods:

• A Bayesian approach is proposed, utilizing Gaussian processes to flexibly model time trends.
• Inference procedures are developed for both sharp and interval null hypotheses.
• The methodology specifically addresses hypotheses testing for group differences at individual time points.

Main Results:

• The proposed Bayesian methodology effectively handles the challenges of highly variable, zero-inflated longitudinal count data.
• Gaussian processes allow for flexible and accurate modeling of time trends.
• The method successfully identified potential wound-healing probiotics in a case study using bacteria counts from mice.

Conclusions:

• The developed Bayesian methodology offers a powerful tool for analyzing complex longitudinal count data, particularly from NGS experiments.
• It provides a flexible and effective way to model time trends and detect group differences.
• The approach is broadly applicable to various biological datasets comparing two subject groups, such as in microbiome or gene expression studies.