PERFect: PERmutation Filtering test for microbiome data.

Area of Science:

• Microbiome research
• Bioinformatics
• Statistical analysis

Background:

• Human microbiota composition is linked to diseases like obesity and inflammatory bowel disease, highlighting its clinical potential.
• Raw microbiome count data present challenges due to sparsity and high dimensionality, requiring robust pre-processing.
• Current filtering standards lack consensus, impacting downstream analysis reproducibility.

Purpose of the Study:

• To introduce PERFect, a novel permutation filtering approach for microbiome data processing.
• To address the need for quantifying filtering loss and assessing excessive filtering.
• To provide a data-driven, statistically principled filtering criterion for microbiome analysis.

Main Methods:

• Developed PERFect, a permutation filtering approach incorporating thresholds to define and quantify filtering loss.
• Implemented a permutation test to evaluate the significance of filtering loss and detect excessive filtering.
• Assessed methods on "mock experiment" datasets with known compositions and two real microbiome datasets.

Main Results:

• PERFect successfully removed contaminant taxa in "mock" datasets and quantified associated filtering loss.
• The method provides uniform, data-driven filtering criteria for real microbiome data.
• In real data analyses, PERFect removed more taxa than existing methods, likely due to its explicit loss function, statistical testing, and consideration of taxa correlations.

Conclusions:

• PERFect offers a statistically sound and data-driven approach to microbiome data filtering.
• The method enhances the reliability and reproducibility of microbiome analyses by addressing filtering standards.
• PERFect provides a valuable tool for researchers seeking to process complex microbiome datasets effectively.