Dissertation Defense: Qingyu Chen

Title: Association Test and Variable Selection in Microbiome Analysis with Phylogeny

Abstract: Microbiome analysis explores the intricate interactions among microbes, hosts, and the environment. The unique features of microbiome data, including compositionality, high-dimensionality, over-dispersion, sparsity, and phylogenetic tree structure, present great challenges that require careful consideration in statistical inference. Our work focuses on two themes: detecting associations between microbial composition and an outcome of interest and identifying influential factors on the human microbiome. Two methods are proposed to tackle these themes. The first method, MiAF, tests an omnibus association between a microbial community and an outcome, balancing OTU-level flexibility with community-level relevance. We combined p-values of OTU-level tests to contruct a community-level test which has considerable statistical power under different microbial profiles. The second project proposes a penalized Dirichlet-tree multinomial mixed-effects regression to select relevant covariates that shape microbiome in longitudinal studies. Splines are used to depict possibly non-linear effect of covariates, especially the trajectory of taxon abundance over time. Our simulation studies demonstrate its potential to capture different underlying association patterns and its robustness under model misspecification.

Zoom link here