Pitfalls in the statistical analysis of microbiome amplicon sequencing data

Area of Science:

• Microbiology
• Bioinformatics
• Statistical Analysis

Background:

• Microbiome data present unique statistical challenges including compositionality, high dimensionality, and excess zeros.
• A wide variety of statistical methods are available, often adapted from other fields like ecology and RNA-sequencing.
• The expanding array of methods necessitates significant researcher effort in selecting appropriate analytical tools.

Purpose of the Study:

• To critically evaluate statistical methods commonly applied to microbiome data.
• To identify and recommend against specific methods that are frequently misused or based on inappropriate assumptions for microbiome datasets.
• To foster a more informed discussion on the suitability of statistical approaches in microbiome research.

Main Methods:

• Review and critical assessment of 14 statistical methods used in microbiome data analysis.
• Evaluation of method assumptions against the characteristics of microbiome data (compositional, high-dimensional, zero-rich).
• Identification of methods applied in ways not intended by their original design.

Main Results:

• A list of 14 statistical methods and approaches recommended for avoidance in microbiome data analysis is presented.
• Specific reasons for avoidance are detailed, including violated statistical assumptions and inappropriate application contexts.
• The study highlights the need for more rigorous critical evaluations of existing statistical tools for microbiome research.

Conclusions:

• Not all statistical methods currently in use are suitable for microbiome data analysis.
• Researchers are advised to exercise caution and critical judgment when selecting statistical methods.
• This work aims to guide researchers towards more appropriate and robust statistical analyses of microbiome data.