Identifying microbial drivers in biological phenotypes with a Bayesian network regression model

Area of Science:

• Microbiome research
• Statistical modeling
• Bioinformatics

Background:

• Bayesian Network Regression (BNR) models are used in brain research to link brain regions to traits.
• Their application to microbiome research, to identify microbial drivers of phenotypes, is unexplored.
• Microbial networks present challenges like high dimensionality and sparsity, differing from brain networks.

Purpose of the Study:

• To investigate the suitability of BNR models for microbial datasets.
• To assess if BNR models, focusing on interaction effects, can identify key microbial drivers of phenotypic variability.
• To provide practical advice for applying BNR models in microbiome research.

Main Methods:

• Evaluation of BNR models using synthetic and real microbial data across diverse biological scenarios.
• Testing the model's ability to identify influential nodes and edges in microbial networks.
• Development of an accessible Julia package for BNR model implementation.

Main Results:

• BNR models successfully identified influential microbial nodes and edges driving phenotypic changes in most tested scenarios.
• Specific scenarios were identified where the BNR model performed poorly, highlighting limitations.
• The study provides practical guidance for domain scientists using BNR for microbiome data.

Conclusions:

• BNR models offer a viable framework for microbiome researchers to uncover connections between microbes and phenotypes.
• The developed Julia package facilitates the application of BNR models in microbiome studies.
• Understanding model limitations is crucial for effective application in microbial ecology.