Testing in Microbiome-Profiling Studies with MiRKAT, the Microbiome Regression-Based Kernel Association Test.

Area of Science:

• Microbiome research
• Statistical genetics
• Computational biology

Background:

• High-throughput sequencing enables human microbiome studies in disease.
• Distance-based methods are popular but face challenges in distance selection and outcome extension.
• Existing methods struggle with choosing the optimal phylogenetic distance and adapting to diverse outcome types.

Purpose of the Study:

• To introduce the Microbiome Regression-based Kernel Association Test (MiRKAT) for robust microbiome association analysis.
• To address limitations of traditional distance-based methods in microbiome research.
• To provide a flexible framework for analyzing microbiome data with various outcomes and covariates.

Main Methods:

• MiRKAT utilizes semi-parametric kernel machine regression to directly model outcome-microbiome associations.
• It employs a variance-component score statistic for association testing and analytical p-value calculation.
• The method supports covariate adjustment and non-parametric microbiome modeling using phylogenetic distances.

Main Results:

• MiRKAT demonstrates correctly controlled Type I error rates and adequate statistical power in simulations.
• "Optimal" MiRKAT, using multiple distances, shows robustness and significant power gains compared to single-distance approaches.
• Application to real data confirmed associations between microbial communities and smoking/fecal protease levels.

Conclusions:

• MiRKAT offers a powerful and flexible alternative to distance-based methods for microbiome-wide association studies.
• The "optimal" MiRKAT approach effectively handles the challenge of selecting appropriate distance metrics.
• This method advances the analysis of microbiome data, revealing significant associations with environmental exposures and host factors.