Statistical methods for testing differential gene expression in bulk and single-cell RNA sequencing data

The transcriptome is the complete set of RNA molecules in biological samples, and they are primarily invariant in different tissues and cells of the same individual. Studying the transcriptome enables understanding of the genome function in response to biological or experimental factors, such as disease and development. RNA sequencing (RNA-seq) uses massively parallel sequencing technologies to profile the transcriptome. Among others, the objective of many RNA-seq studies is to identify features of the transcriptome (such as genes) that are deferentially expressed (DE) between or among groups of individuals or tissues that are different with respect to some condition (e.g. disease status and treatment). However, because of technical artefacts in RNA-seq technologies and the dynamics of the biological system, the profiling (quantifying) of the transcriptome is typically subjected to an inherent variation even within the same condition. Consequently, statistical methods are principled approaches that help biologists understand to what extent (on average) genes are DE using RNA-seq datasets. In this doctoral thesis, challenges associated with DE analysis of RNA-seq data are explored, and various solutions have been proposed. In particular, existing statistical methods for DE analysis are comprehensively evaluated, a novel simulation tool for RNA-seq data is proposed to facilitate realistic evaluation of statistical methods for DE analysis, a new method is proposed to improve the existing distribution-free methods for DE analysis, and cost-effective experimental designs for RNA-seq studies are explored. The findings discussed in this dissertation are relevant to all scientists and clinicians involved in RNA-seq studies, in particular for testing DE, benchmarking statistical and bioinformatics tools, designing cost-effective RNA-seq experiments