Advancement: Bayesian Modeling for Count Data with Applications to Microbiome Studies

Speaker Name: 
Kurtis Shuler
Speaker Title: 
PhD Student
Speaker Organization: 
Statistics & Applied Mathematics
Start Time: 
Wednesday, May 30, 2018 - 10:00am
End Time: 
Wednesday, May 30, 2018 - 12:00pm
Location: 
Baskin Engineering, Room 330
Organizer: 
Juhee Lee

Abstract:  High throughput sequencing technology has paved the way for microbiome analyses, leading to new insights into microbial communities and their interaction with environments. Despite advances in sequencing technology, modeling how microbiomes and environmental factors interact remains challenging. Count data produced by sequencing is typically high-dimensional and over-dispersed, and controlling for different library sizes in microbiome samples is not always straightforward. In this proposal we develop Bayesian statistical models that handle these complications and address questions oftentimes asked by biologists. Specifically, we include 1) A sparse regression model with a non-local prior to identify important factors related to microbiome abundances, 2) A nonparametric model for comparing microbial abundance distributions, and 3) A graphical model with nonlinear regression to account for interactions among the organisms present in mi! crobiomes. We show through simulation studies that the proposed approaches provide improved inference over existing methods, and then we apply the models to two real world microbiome datasets. The first analysis uses ocean microbiome data to study how microbial abundance levels relate to environmental factors in the context of harmful algal blooms, and the second considers the microbial communities in Daphia Magna with different immunity traits.