518 – Statistical Methods for Complex Interactions and Genetic and Environmental Epidemiology
Bayesian Sparse Functional Principal Components Analysis Models Dynamic Temporal Changes in Longitudinal Microbiome Studies
Antonio Gonzalez
University of California San Diego
Lingjing Jiang
University of California San Diego
Rob Knight
University of California San Diego
Loki Natarajan
University of California San Diego
Wesley K. Thompson
University of California San Diego
Yoshiki Vazquez-Baeza
University of California San Diego
Longitudinal microbiome studies provide valuable information about dynamic interactions between the microbiome and host, by capturing both between-individual differences and within-subject dynamics. As the costs of DNA sequencing have decreased, microbiome researchers have a greater opportunity to perform longitudinal studies to better understand microbial changes in response to an intervention in real time. However, microbial communities can change abruptly in response to small perturbations. Current approaches for longitudinal microbiome analysis are not sufficient to capture this dynamic temporal variation, especially with the additional challenges of irregular sampling intervals, limited sample size, missing values and dropouts. We developed a Bayesian Sparse Functional Principal Components Analysis (SFPCA) methodology to meet the growing need to model dynamic temporal change and to detect its dependence on biological covariates in longitudinal microbiome analysis. We show in simulations and in real data applications that Bayesian SFPCA is able to overcome the above challenges in longitudinal microbiome analysis, and is more sensitive for capturing temporal variations and detecting differences due to biological covariates than existing methods. We therefore expect Bayesian SFPCA to be a valuable tool to the microbial community for longitudinal analysis.