TY - JOUR
T1 - A Bayesian Negative Binomial Hierarchical Model for Identifying Diet–Gut Microbiome Associations
AU - Revers, Alma
AU - Zhang, Xiang
AU - Zwinderman, Aeilko H.
N1 - Publisher Copyright: © Copyright © 2021 Revers, Zhang and Zwinderman.
PY - 2021/10/7
Y1 - 2021/10/7
N2 - The human gut microbiota composition plays an important role in human health. Long-term diet intervention may shape human gut microbiome. Therefore, many studies focus on discovering links between long-term diets and gut microbiota composition. This study aimed to incorporate the phylogenetic relationships between the operational taxonomic units (OTUs) into the diet-microbe association analysis, using a Bayesian hierarchical negative binomial (NB) model. We regularized the dispersion parameter of the negative binomial distribution by assuming a mean-dispersion association. A simulation study showed that, if over-dispersion is present in the microbiome data, our approach performed better in terms of mean squared error (MSE) of the slope-estimates compared to the standard NB regression model or a Bayesian hierarchical NB model without including the phylogenetic relationships. Data of the Healthy Life in an Urban Setting (HELIUS) study showed that for some phylogenetic families the (posterior) variances of the slope-estimates were decreasing when including the phylogenetic relationships into the analyses. In contrast, when OTUs of the same family were not similarly affected by the food item, some bias was introduced, leading to larger (posterior) variances of the slope-estimates. Overall, the Bayesian hierarchical NB model, with a dependency between the mean and dispersion parameters, proved to be a robust method for analyzing diet-microbe associations.
AB - The human gut microbiota composition plays an important role in human health. Long-term diet intervention may shape human gut microbiome. Therefore, many studies focus on discovering links between long-term diets and gut microbiota composition. This study aimed to incorporate the phylogenetic relationships between the operational taxonomic units (OTUs) into the diet-microbe association analysis, using a Bayesian hierarchical negative binomial (NB) model. We regularized the dispersion parameter of the negative binomial distribution by assuming a mean-dispersion association. A simulation study showed that, if over-dispersion is present in the microbiome data, our approach performed better in terms of mean squared error (MSE) of the slope-estimates compared to the standard NB regression model or a Bayesian hierarchical NB model without including the phylogenetic relationships. Data of the Healthy Life in an Urban Setting (HELIUS) study showed that for some phylogenetic families the (posterior) variances of the slope-estimates were decreasing when including the phylogenetic relationships into the analyses. In contrast, when OTUs of the same family were not similarly affected by the food item, some bias was introduced, leading to larger (posterior) variances of the slope-estimates. Overall, the Bayesian hierarchical NB model, with a dependency between the mean and dispersion parameters, proved to be a robust method for analyzing diet-microbe associations.
KW - Bayesian
KW - associations
KW - diet
KW - hierarchical model
KW - microbiome data
UR - http://www.scopus.com/inward/record.url?scp=85117476992&partnerID=8YFLogxK
U2 - https://doi.org/10.3389/fmicb.2021.711861
DO - https://doi.org/10.3389/fmicb.2021.711861
M3 - Article
C2 - 34690956
SN - 1664-302X
VL - 12
JO - Frontiers in Microbiology
JF - Frontiers in Microbiology
M1 - 711861
ER -