Comprehensive evaluation of statistical approaches for differential metaproteomics

Area of Science:

• Microbiology
• Bioinformatics
• Proteomics

Background:

• Metaproteomics analyzes protein expression in microbial communities.
• Statistical methods are crucial but face challenges like data sparsity and variability.
• Current understanding of optimal statistical approaches for metaproteomics is limited.

Purpose of the Study:

• To identify the most effective data processing methods and statistical tests for differential protein analysis in metaproteomics.
• To establish a framework for evaluating statistical approaches using defined metaproteomic datasets.

Main Methods:

• Generated 13 metaproteomic samples with known compositions and differences.
• Compared over 110 statistical analysis combinations, including regression, inference, and machine learning.
• Utilized raw mass spectrometry data and provided reproducible benchmarking code.

Main Results:

• Several statistical method combinations demonstrated suitability for differential expression analysis.
• Effective approaches were identified within limma, edgeR, MaAslin2, linear models, and random forests.
• Key recommendations for metaproteomic differential expression analysis were highlighted.

Conclusions:

• The study provides a framework for assessing statistical methods in metaproteomics.
• Specific combinations of data processing and statistical tests can reliably identify differentially abundant proteins.
• This work improves the assessment of statistical methods for metaproteomic data analysis.