A distance based multisample test for high-dimensional compositional data with applications to the human microbiome.

Area of Science:

• Statistics
• Bioinformatics
• Genomics

Background:

• Compositional data, common in genomics and geology, sum to one, requiring specialized statistical methods.
• Traditional statistical tests are unsuitable for compositional data due to constraints and potential over-dispersion.
• Analysis of high-dimensional, over-dispersed compositional data, such as microbiome data, is challenging.

Purpose of the Study:

• To develop a novel statistical test for detecting compositional differences between multiple populations.
• To address the limitations of existing methods for analyzing complex, high-dimensional compositional data.
• To provide a robust and efficient method for microbiome and metagenomics research.

Main Methods:

• A Bayesian hypothesis formulation for compositional difference testing.
• A nonparametric test utilizing inter-point distances for statistical significance.
• Direct analysis of compositional data without transformations or restrictive assumptions.

Main Results:

• The proposed test effectively identifies compositional differences in simulated and real microbiome data.
• The method demonstrates higher sensitivity compared to mean-based approaches, especially for over-dispersed or zero-inflated data.
• The test is robust, requiring no data transformation, sparsity assumption, or specific covariance matrix conditions.

Conclusions:

• The developed nonparametric test is a sensitive and efficient tool for analyzing compositional differences.
• Its ease of implementation and computational efficiency make it suitable for large-scale microbiome and metagenomics datasets.
• The method offers a valuable alternative for researchers dealing with complex compositional data structures.