B0561
Title: Covariate-adjusted Bayesian kernel regression for learning effect sizes of selected microbiome
Authors: Liangliang Zhang - Case Western Reserve University (United States) [presenting]
Christine Peterson - The University of Texas MD Anderson Cancer Center (United States)
Abstract: Current microbiome profiling methods allow for very fine resolution of the strains present in each sample. When associated with patient-level outcomes, the abundant features tend to be more influential than the rare features. Therefore, we propose a Bayesian kernel method to convolute all the microbial features together and study their joint impact in nonlinear kernel regression. The method can balance the impact of abundant features and rare features by taking into account their internally linked kinship structures and provide a similarity function which further helps in categorizing patient groups. Unlike the linear regression setting, there is no clear form of effect sizes in kernel regression. We transform the Kernel space back to the original space to obtain estimated effect sizes for individual microbiome features, which is usually not an easy task to accomplish. In addition, the model will provide us with improved uncertainty assessment both at the joint level and the individual level.