llperm: a permutation of regressor residuals test for microbiome data

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Modeling

Background:

• Differential abundance testing is crucial for microbiome analysis.
• Existing models often fail to fit microbiome data, leading to high false positive rates.
• Permutation tests are alternatives, but regression approaches are needed for complex datasets.

Purpose of the Study:

• To introduce a robust statistical method for microbiome differential abundance testing.
• To address limitations of standard regression models in microbiome analysis.
• To provide a reliable tool for analyzing small microbiome datasets with multiple covariates.

Main Methods:

• Implementation of the Permutation of Regressor Residuals (PRR) test in an R package named 'llperm'.
• The PRR test is applicable to any likelihood-based model, including those handling zero-inflation and overdispersion.
• Utilized simulations based on a real microbiome dataset for evaluation.

Main Results:

• The PRR-test maintains the nominal false positive rate under the null hypothesis.
• It demonstrates equal or greater power in detecting true positives compared to standard likelihood-based models.
• The method is suitable for count regression models commonly used in microbiome research.

Conclusions:

• Standard count regression models exhibit high false positive rates in microbiome data.
• The PRR-test offers a guaranteed nominal false positive rate, preventing erroneous conclusions.
• This approach enhances the reliability of microbiome data analysis.