Benchmarking differential abundance analysis methods for correlated microbiome sequencing data

Area of Science:

• Microbiome research
• Statistical analysis
• Bioinformatics

Background:

• Differential abundance analysis (DAA) is crucial for identifying microbial biomarkers.
• Microbiome studies often involve correlated samples (spatial, temporal).
• Existing DAA tools for correlated data (DAA-c) produce inconsistent results.

Purpose of the Study:

• To comprehensively evaluate existing DAA-c tools.
• To identify the most robust and accurate DAA-c methods.
• To provide best practice recommendations for microbiome data analysis.

Main Methods:

• Real data-based simulations were used for evaluation.
• Performance of various DAA-c tools was compared.
• Methods included linear models (LinDA, MaAsLin2, LDM) and generalized linear models.

Main Results:

• Linear model-based methods (LinDA, MaAsLin2, LDM) demonstrated greater robustness.
• LinDA maintained performance even with strong compositional effects.
• Significant discordance was observed among different DAA-c tools.

Conclusions:

• Linear model-based approaches are recommended for DAA in correlated microbiome data.
• LinDA is highlighted as a particularly reliable tool, especially for compositional data.
• Standardized evaluation is needed to guide the selection of DAA-c tools.