Combining p-values from various statistical methods for microbiome data

Area of Science:

• Microbiome analysis
• Statistical genetics
• Bioinformatics

Background:

• Microbiome data is characterized by overdispersion and sparsity.
• Existing statistical methods for identifying differential features yield inconsistent results.
• Integrating results from multiple methods is essential for robust microbiome taxa importance determination.

Purpose of the Study:

• To evaluate and identify the most suitable meta-analysis method for microbiome data.
• To compare the performance of various meta-analysis techniques in assessing microbiome taxa significance.

Main Methods:

• Evaluated Fisher's method, minimum p-value, Simes, Stouffer's, Kost method, and Cauchy combination test.
• Conducted simulation studies to assess method performance and type 1 error rates.
• Applied methods to colorectal cancer microbiome data for real-world validation.

Main Results:

• The Cauchy combination test demonstrated superior performance in combining p-values.
• It effectively controlled type 1 error rates and showed high rank similarity with true ranks in simulations.
• Application to colorectal cancer data identified known associated taxa, validating its efficacy.

Conclusions:

• The Cauchy combination test is the recommended meta-analysis method for microbiome data.
• This method provides a reliable approach for determining the importance of microbiome taxa.
• It enhances the interpretability and reliability of microbiome association studies.