BCGLMs: Bayesian modeling for disease prediction using compositional microbiome features

Area of Science:

• Bioinformatics
• Computational Biology
• Statistical Modeling

Background:

• Compositional data analysis is crucial for understanding complex biological systems like the microbiome.
• Existing methods may not fully capture the nuances of microbiome data, such as phylogenetic relationships and random effects.
• Bayesian approaches offer a flexible framework for modeling complex data structures.

Purpose of the Study:

• To introduce BCGLMs, a novel R package for Bayesian compositional data analysis.
• To provide tools for fitting models with various response types and incorporating random effects.
• To enable the integration of phylogenetic information into microbiome data modeling.

Main Methods:

• Development of the BCGLMs R package, built upon the brms package.
• Implementation of functions for setting up and fitting Bayesian compositional generalized linear models (BCGLMs).
• Inclusion of capabilities for handling continuous, binary, ordinal, and survival responses.
• Integration of random effects for improved prediction accuracy.
• Facilitation of phylogenetic relationship incorporation for microbiome taxa.

Main Results:

• BCGLMs offers a comprehensive suite of tools for Bayesian compositional data analysis.
• The package supports diverse response variables and advanced modeling techniques.
• Users can leverage phylogenetic information for more accurate microbiome analysis.
• Tools for both numerical and graphical summarization of model results are provided.

Conclusions:

• BCGLMs provides a flexible and powerful framework for analyzing compositional microbiome data.
• The package enhances prediction accuracy through the inclusion of random effects and phylogenetic relationships.
• BCGLMs democratizes advanced Bayesian modeling for microbiome research.