Novel Methods For Next-Generation Sequencing Data With Applications in Microbiome Studies

Abstract

Humans maintain a symbiotic relationship with the billions of microbes that exist within and upon the body. The collection of microbes within the body can be considered as a second genome, providing a plethora of unique information about their host. High-throughput next generation sequencing technologies have allowed researchers to build microbial profiles based on microbial RNA sequences for individuals/patients, providing a rich avenue of data to be utilized in statistical models in various field including medicine and forensics. In this dissertation, I present three novel projects which utilize next-generation sequencing based microbiome profiles. In the first project, I proposed a new approach based on microbiome dissimilarity measurements, with applications in forensic trace evidence analysis. This approach utilizes bootstrap Aitchison distances between communities to identify groups of microbial samples and improve current source tracking applications for evidence analysis by removing samples that are highly dissimilar to the evidence. The last two projects focus on detection of mediation effects when the microbes are treated as mediators in clinical mediation models. The second project aims to identify mediation of immune response genes on human gut inflammation non-parametrically by applying information theory concepts from machine learning. The third project expands the scope of mediation modeling by considering time series data in conjunction with mediation. Comprehensive simulation experiments show a drastic improvement in the detection of mediation effects compared to current standard methods for models which utilize microbiome as a mediator.