A1884
Title: Bayesian sparse principal coordinates analysis with microbiome discoveries
Authors: Shao-Hsuan Wang - National Central University (Taiwan) [presenting]
Abstract: Principal coordinates analysis (PCoA) is a widely used tool for exploring relationships among samples based on pairwise dissimilarities. However, classical PCoA typically produces dense loading structures, which makes interpretation difficult, particularly in ultrahigh-dimensional microbiome studies where identifying biologically relevant features is essential. A sparse PCoA framework that incorporates regularization to produce interpretable coordinate axes with explicit feature selection is introduced. To further accommodate flexible sparsity structures, a Bayesian sparse PCoA model is developed that employs global-local shrinkage priors from the three-parameter beta-normal family to achieve adaptive sparsity. Asymptotic results clarify how sample size and dimensionality influence the behavior of PCoA. Through simulation studies and analyses of the Hadza gut microbiome and pancreatic cancer tumor microbiome datasets, the proposed method substantially improves sparsity and interpretability while preserving key ecological and clinical signals.
