Structured variable selection with q-values.

Area of Science:

• Metabolomics
• Microbiome research
• Statistical genetics

Background:

• Variable selection is challenging when covariates influence both responses and other covariates, especially with more variables than samples.
• Metabolic studies in mice, involving diet and gut microbial composition, present such complex data structures.
• Understanding diet's direct effects on phenotypes and microbial percentages is crucial for accurate analysis.

Purpose of the Study:

• To develop a novel methodology for variable selection in high-dimensional datasets with complex covariate relationships.
• To identify specific gut microbial taxa that influence host phenotypes, such as body weight regulation.
• To account for the direct impact of dietary interventions on both host phenotypes and microbial communities.

Main Methods:

• A new statistical approach is presented, integrating q-values from multiple hypothesis testing.
• The methodology utilizes the recently developed weighted Lasso technique for variable selection.
• This approach is applied to a metabolic study dataset from mice with dietary interventions and microbiome data.

Main Results:

• The proposed method effectively performs variable selection in scenarios with intercorrelated regressors.
• It successfully identifies microbial features associated with body weight regulation phenotypes.
• The methodology accounts for the dual role of diet as a direct influence and a confounder.

Conclusions:

• The novel weighted Lasso approach provides a robust framework for variable selection in complex biological systems.
• This method enhances our ability to pinpoint microbial contributions to host phenotypes amidst confounding dietary effects.
• The findings have implications for understanding host-microbiome-diet interactions in metabolic health.