Year of Publication


Degree Name

Doctor of Philosophy (PhD)

Document Type

Doctoral Dissertation




Computer Science

First Advisor

Dr. Zongming Fei

Second Advisor

Dr. Licong Cui


Cancer is a genetic disease responsible for one in eight deaths worldwide. The advancement of next-generation sequencing (NGS) technology has revolutionized the cancer research, allowing comprehensively profiling the cancer genome at great resolution. Large-scale cancer genomics research has sparked the needs for efficient and accurate Bioinformatics methods to analyze the data. The research presented in this dissertation focuses on three areas in cancer genomics: cancer somatic mutation detection; cancer driver genes identification and transcriptome profiling on single-cell level.

NGS data analysis involves a series of complicated data transformation that convert raw sequencing data to the information that is interpretable by cancer researchers. The first project in the dissertation established a robust, reproducible and scalable cancer genomics data analysis workflow management system that automates the best practice mutation calling pipelines to detect somatic single nucleotide polymorphisms, insertion, deletion and copy number variation from NGS data. It integrates mutation annotation, clinically actionable therapy prediction and data visualization that streamlines the sequence-to-report data transformation.

In order to differentiate the driver mutations buried among a vast pool of passenger mutations from a somatic mutation calling project, we developed MEScan in the second project, a novel method that enables genome-scale driver mutations identification based on mutual exclusivity test using cancer somatic mutation data. MEScan implements an efficient statistical framework to de novo screen mutual exclusive patterns and in the meantime taking into account the patient-specific and gene-specific background mutation rate and adjusting the heterogenous mutation frequency. It outperforms several existing methods based on simulation studies and real-world datasets. Genome-wide screening using existing TCGA somatic mutation data discovers novel cancer-specific and pan-cancer mutually exclusive patterns.

Bulk RNA sequencing (RNA-Seq) has become one of the most commonly used techniques for transcriptome profiling in a wide spectrum of biomedical and biological research. Analyzing bulk RNA-Seq reads to quantify expression at each gene locus is the first step towards the identification of differentially expressed genes for downstream biological interpretation. Recent advances in single-cell RNA-seq (scRNA-seq) technology allows cancer biologists to profile gene expression on higher resolution cellular level. Preprocessing scRNA-seq data to quantify UMI-based gene count is the key to characterize intra-tumor cellular heterogeneity and identify rare cells that governs tumor progression, metastasis and treatment resistance. Despite its popularity, summarizing gene count from raw sequencing reads remains the one of the most time-consuming steps with existing tools. Current pipelines do not balance the efficiency and accuracy in large-scale gene count summarization in both bulk and scRNA-seq experiments. In the third project, we developed a light-weight k-mer based gene counting algorithm, FastCount, to accurately and efficiently quantify gene-level abundance using bulk RNA-seq or UMI-based scRNA-seq data. It achieves at least an order-of-magnitude speed improvement over the current gold standard pipelines while providing competitive accuracy.

Digital Object Identifier (DOI)