QBioS Thesis Defense, Conan Zhao

In partial fulfillment of the requirements for the degree of  
Doctor of Philosophy in Quantitative Biosciences
in the School of Biological Sciences

Conan Zhao

Defends his thesis:

Identifying Microbial Biomarkers of Cystic Fibrosis Health and Disease

Tuesday, June 21, 2022
10am Eastern Time
In-person: Marcus Nanotechnology Building, Room #1117-1118
Virtual: https://gatech.zoom.us/j/2903027082 
Open to the Community

Dr. Sam Brown
School of Biological Sciences
Georgia Institute of Technology

Committee Members:
Dr. Joshua Weitz
School of Biological Sciences, School of Physics
Georgia Institute of Technology 

Dr. Arlene Stecenko
Department of Pediatrics
Emory University School of Medicine

Dr. Peng Qiu
School of Biomedical Engineering
Georgia Institute of Technology / Emory University

Dr. Rishi Kamaleswaran
Department of Biomedical Informatics
Emory University

Chronic, polymicrobial respiratory infections remain the primary driver of morbidity and mortality in cystic fibrosis (CF). This thesis leverages experimental data and large-scale public datasets to investigate the relationships between microbiome structure, pathogen abundance and host health. 

First, using a machine learning framework, we show that off-the-shelf machine learning methods can recover known clinical and microbial predictors of lung function from a set of 77 sputum composition profiles. These methods recover known demographic predictors of lung function and further identify novel taxonomic predictors, highlighting the utility of simple machine learning methods for microbial biomarker discovery. 

Second, we develop a synthetic infection microbiome model representing CF metacommunity diversity, and benchmark on clinical data.  Using this synthetic microbiome system, we provide evidence that commonly used CF antibiotics can drive the expansion (via competitive release) of previously rare opportunistic pathogens and offer a path towards microbiome-informed treatment strategies. 

Last, we manually curated a microbiome dataset of over 4000 sputum samples representing more than 1000 people with CF (pwCF), matching samples with corresponding metadata from 36 publications and standardizing bioinformatic analyses with a single common pipeline. We fit Sloan Neutral Community Models to each study and find a consistent set of neutral and non-neutral taxa.  Using Dirichlet Multinomial Mixture modeling, we partition non-neutral CF lung microbiomes into 14 distinct pulmotypes. Integrating longitudinal data, we find that not all Pseudomonas-dominated pulmotypes are dynamically equivalent, which carries important implications for infection management in cystic fibrosis.