CMStatistics 2018: Start Registration
View Submission - CMStatistics
B1030
Title: Multi-task Dirichlet-multinomial regression for detecting global microbiome associations Authors:  Frank Dondelinger - Lancaster University (United Kingdom) [presenting]
Abstract: There is evidence that the human gut microbiome influences diseases as disparate as inflammatory bowel disease, cardiovascular disease and schizophrenia. Current statistical techniques for microbiome association studies either rely on a measure of microbiome distance, or on detecting associations with individual bacterial species. A method that extends the latter approach beyond individual species is the multi-task Dirichlet-multinomial model; however, it does not take species relatedness into account. We have improved on that approach in two respects: 1) by incorporating the phylogenetic tree of the microbial species as prior information about their relatedness, and 2) by introducing a hierarchical Bayesian prior that allows us to estimate the global effect of each covariate on the microbiome. We have applied our method to simulated data, and show that it allows for better estimation of global effects compared to a post-processing of the individual effects detected by the previous method. Additionally, we apply the data to two real-world examples in Crohn's disease and in inflammatory bowel disease. We show that our method can reliably detect global associations that are supported by the literature.