A nutritionally informed model for Bayesian variable selection with metabolite response variables

Area of Science:

• Nutritional Epidemiology
• Metabolomics
• Statistical Genetics

Background:

• Diet significantly impacts human metabolism, but identifying specific food-metabolite associations is challenging due to complex data.
• Existing statistical methods struggle with high-dimensional metabolomic data exhibiting skewness, censoring, and missingness.

Purpose of the Study:

• To develop a novel statistical framework for identifying associations between food intake and blood metabolites.
• To address the statistical complexities inherent in metabolomic data and high-dimensional variable selection.

Main Methods:

• A Bayesian variable selection framework using a skew-normal censored mixture model was developed.
• A Markov random field prior was employed to incorporate nutritional and statistical relationships among food intake variables.
• The methodology was applied to data from two healthcare professional cohorts.

Main Results:

• The novel approach identified multiple significant associations between food items and blood metabolites.
• These findings include associations consistent with prior research and several potentially new discoveries.
• The method outperformed standard approaches in detecting these metabolite-diet links.

Conclusions:

• The proposed Bayesian framework effectively identifies diet-metabolite associations in high-dimensional metabolomic data.
• The R package multimetab provides a practical tool for researchers in nutritional epidemiology.
• This methodology advances our understanding of how diet influences human metabolism.