Human microbiome research seeks to better understand the role of our microbial communities and how they interact with their host, respond to their environment and influence disease. \Recently published in the Annals of Applied Statistics, Matt Koslovsky, an assistant professor in the Department of Statistics, and colleagues developed a Bayesian joint model that simultaneously identifies clinical covariates associated with microbial composition data and predicts a phenotypic response using information contained in the compositional data. Previous efforts have modeled these data separately, which can produce biased interpretations and reduce prediction performance. By integrating these data together, their work may help researchers design interventions that modulate the composition of the microbiome to promote health and to cure disease.
Microbiome data are challenging to model in part due to their compositional structure and high-dimensionality. In practice, the computational complexity of regression models designed for microbiome data analysis is compounded by large covariate spaces. To address this, the authors introduce a novel data augmentation strategy for compositional data and embed sparsity inducing priors at both levels of the model to improve the interpretability of the results. Additionally, they provide a user-friendly R package, MicroBVS, that can accommodate alternative model and prior specifications for microbiome data analysis. Their work will be presented at the “Best of Annals of Applied Statistics” session this summer at the Joint Statistical Meetings.