A0222
Title: Structured Bayesian variable selection for microbiome compositional data using graph-guided shrinkage
Authors: Satabdi Saha - University of Texas MD Anderson Cancer Center (United States) [presenting]
Christine Peterson - The University of Texas MD Anderson Cancer Center (United States)
Abstract: A Bayesian regression model for microbiome compositional data that accounts for both sparsity and microbial structure is proposed. The method employs a structured horseshoe prior that encourages variable selection while borrowing strength across related taxa through a graph-informed shrinkage. To respect compositional constraints, regression coefficients are modeled under a sum-to-zero condition. Posterior inference is performed via an efficient blocked Gibbs sampler with Langevin updates. Through simulation studies and application to real microbiome data, improved feature selection and predictive performance are demonstrated over existing state-of-the-art algorithms.
