New statistical method identifies cytokines that distinguish stool microbiomes

Area of Science:

• Microbiology
• Statistical Modeling
• Bioinformatics

Background:

• Regression analysis is crucial for understanding variable associations, but standard methods struggle with complex microbiome taxa count data.
• Existing regression models for microbiome data often require restrictive assumptions, limiting their applicability.

Purpose of the Study:

• To develop a flexible, non-parametric regression model for microbiome taxa count data.
• To provide an interpretable and automatically fitting model for analyzing microbiome composition and associated metadata.

Main Methods:

• The proposed model, Dirichlet-multinomial with recursive partitioning (DM-RPart), combines statistical distributions with tree-based partitioning.
• This approach allows for non-parametric regression on microbiome taxa counts and other compositional data.

Main Results:

• The DM-RPart model was successfully applied to both cytokine and microbiome taxa count data.
• The model demonstrates applicability to diverse microbiome datasets and associated metadata.

Conclusions:

• DM-RPart offers a novel and effective regression framework for microbiome and compositional data analysis.
• The associated R package (HMP) is available on CRAN, facilitating broader adoption and application.