Statistical Modeling of Next Generation Sequencing Data

The advent and development of next generation sequencing (NGS) has transformed our ways to study biological questions. These new technologies have brought new developments in many different areas, such as disease-associated rare variants identification, point and structural variants detection, whole transcriptome studies, and high-resolution target identification of binding proteins. Meanwhile, large-scale datasets generated by next generation sequencing also pose big challenges in experimental designs and data analysis.;To facilitate better understanding of biological questions through next generation sequencing data, we develop statistical and computational methods to model NGS data, including DNA-seq, RNA-seq and CLIP-seq. In this dissertation, we discuss the unwanted variance modeling of pooled DNA-seq and apply the DNA-seq analysis methods to detect cancer-associated variants in microRNAome and 3UTRome; we design a novel statistical approach to jointly analyze multiple expression profiles (by RNA-seq and microarray) to identify the microRNA-gene interaction network in cancers; we also consider the differential expression study with small replicate RNA-seq samples on the application of aging study; we then systematically study the statistical issues of CLIP-seq and apply the findings on the application of LIN-28 protein CLIP-seq datasets.;Our works not only facilitate better understanding of features of next generation sequencing data, but also provide guidance for experimental designs and higher power in data analysis, which finally help decipher the complicated biological questions through these high-throughput large-scale datasets.