Statistical methods for analysis of overdispersed and correlated next-generation sequencing data

Abstract

Advances in next-generation sequencing (NGS) have fundamentally transformed the study of genome biology, providing researchers with an unprecedented ability to investigate gene expression and epigenetic modifications at high resolution. This technological breakthrough has driven advances in multiple “-omics” fields, including transcriptomics, epigenomics, metagenomics, and more. Unlike traditional sequencing methods, NGS allows for parallel sequencing of hundreds to thousands of genes, or even entire genomes, within a short timeframe. With vast amounts of data, NGS offers insights into fundamental biological processes and their role in disease. However, the complexity of NGS data presents significant statistical challenges, requiring specialized methods to account for overdispersed and correlated data, and ambiguities in read assignment. The main goal of this dissertation was the development of statistical methodologies tailored for the analysis of epigenomic and transcriptomic datasets, which are often characterized by complex structures, including overdispersion, correlation, and dependence between observations. These challenges arise due to the nature of bisulfite sequencing (BS-seq) and RNA sequencing (RNA-seq) technologies, both of which generate high-dimensional and structured count-based data requiring advanced statistical approaches for accurate inference.