Hypothesis Testing and Power Analysis for Longitudinal Microbiome Experiments

• Activity Number: 294
• Type: Contributed
• Date/Time: Tuesday, July 31, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Patricio Salvatore La Rosa*+ and Elena Deych and Berkley Shands and Yanjiao Zhou and Erica Sodergren and George Weinstock and William Shannon
• Companies: Washington University in St. Louis School of Medicine and Washington University School of Medicine and Washington University in St. Louis School of Medicine and Washington University in St. Louis School of Medicine and Washington University in St. Louis School of Medicine and Washington University in St. Louis School of Medicine and Washington University School of Medicine
• Keywords: Human Microbiome Data ; Hypothesis testing ; Microbial taxa abundance distribution ; Generalized Dirichlet Multinomial ; Power Analysis

Abstract

Human microbiome research uses next generation sequencing to characterize the human-associated microbial population and its impact on human health. In recent years much progress has been achieved in optimizing the processes for collecting microbiome samples, processing the DNA, and generating taxonomies/phylogenies from these sequences, however, there are few formal methods for designing and analyzing these experiments, with most approaches being applicable to the particular problem being faced. In this work we develop statistical methods to model taxa abundance distribution of bacteria and to perform hypothesis testing. In particular, we apply the generalized Dirichlet Multinomial model of Wilson to model varying taxa abundance distribution of bacteria over time. This model accounts for the extra variation across subjects and within subjects, as well as incorporating variable sequence depth per subject. Using this model we develop tests of hypotheses about one-group and two-group comparisons, and study their power and size under sample size scenarios (sequence depth and subjects) typical of HMP data. As an example, we apply this methodology to analyze existing HMP data.