Novel Statistical Methods for Detection and Interpretation of Cancer Biomarkers

Cancer biomarkers are molecular signals detected in blood or other human tissues that are indicative of tumor development status or treatment responses. With the recent advances of high throughput technologies, high dimensional molecular profiles are generated with the potential of identifying and interpreting biomarker signals from them.;The statistical methodologies developed in this thesis are motivated by our interest in studying cancer biomarkers using either gene expression profiles or the DNA sequencing profiles. In Chapter 2 and 3, we consider two novel directions of assembling the RNA level gene expression signals. In Chapter 4, we consider the novel idea of 'liquid biopsy' and develop a new method to identify somatic mutation biomarkers in tumor DNAs sampled in the blood.;Specifically, in Chapter 2, we analyzed the TCGA Pan-Cancer gene expression data and demonstrated the surprising result that gene expression signals in adjacent normal samples are more predictive of patients' survival than in tumor samples themselves. Our results suggest the potential benefit of collecting and profiling matched normal tissues to gain more insights on disease etiology and patient progression. In Chapter 3, we used the qPCR data of macrophage expression signals and assemble the signals into an anti-tumor metric as the biomarker. We implemented an EM algorithm to jointly estimate the Ml/M2 ratio based on the gene expression of M1 and M2 types of macrophage genes. In Chapter 4, we used the DNA sequencing data of circulating tumor DNAs (ctDNAs) to identify possible somatic mutation biomarkers in the circulation system. We develop a novel statistical methodology that is more sensitive and specific to call somatic mutations in ctDNAs and conveniently generates a metric to estimate the overall tumor load of the sample. In Chapter 5, we provide a summary of the study and directions for future research.;We hope our work will facilitate better understanding of cancer biomarkers and inspire novel biomarkers being put into practical use to benefit cancer patients in the real world.