MPrESS: An R-Package for Accurately Predicting Power for Comparisons of 16S rRNA Microbiome Taxa Distributions...

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Modeling

Background:

• Human microbiome composition varies between populations, detectable via 16S rRNA gene sequencing.
• Limited sample sizes can hinder microbiome study power.
• Dirichlet mixture modeling (DMM) can simulate microbiome data when experimental data is insufficient.

Purpose of the Study:

• To evaluate the accuracy of DMM-simulated microbiome data in reflecting experimental diversity and calculating statistical power.
• To assess the performance of DMM in power estimation for 16S rRNA gene studies.
• To introduce an R-package, MPrESS, for microbiome power and sample size calculations.

Main Methods:

• Comparison of DMM-simulated 16S rRNA gene data with experimental microbiome data.
• Analysis of power overestimation by DMM under varying data discrepancies.
• Evaluation of DMM admixtures with experimental data.
• Development and introduction of the MPrESS R-package.

Main Results:

• DMM consistently overestimates statistical power, even with <10% data difference, except for highly discriminating taxa.
• DMM admixtures with experimental data performed poorly and lacked correlation with experimental p-values and power.
• Pure DMM simulation showed limitations in accurately reflecting experimental data diversity and power.

Conclusions:

• While DMM simulation can be useful when sample size exceeds requirements, it generally overestimates power.
• Multiple random sampling replications remain the preferred method for determining statistical power.
• The MPrESS R-package provides a tool for accurate power and sample size estimation in 16S rRNA gene microbiome studies.