published_or_final_version / abstract / toc / Clinical Oncology / Doctoral / Doctor of Philosophy

Microarray expression data which contains the expression levels of a large number of simultaneously observed genes have been used in many scientific research and clinical studies. Due to its high dimensionalities, selecting a small number of genes has shown to be beneficial for many tasks such as building prediction models from the microarray expression data or gene regulatory network discovery. Traditional gene selection methods, however, fail to take the class distribution into the selection process. In biomedical science, it is very common to have microarray expression data which is severely biased with one class of examples (e.g., diseased samples) significantly less than other classes (e.g., normal samples). These sample sets with biased distributions require special attention from researchers for identification of genes responsible for a particular disease. In this thesis, we propose three filtering techniques, Higher Weight ReliefF, ReliefF with Differential Minority Repeat and ReliefF with Balanced Minority Repeat to identify genes responsible for fatal diseases from biased microarray expression data. Our solutions are evaluated on five well-known microarray datasets, Colon, Central Nervous System, DLBCL Tumor, Lymphoma and ECML Pancreas. Experimental comparisons with the traditional ReliefF filtering method demonstrate the effectiveness of the proposed methods in selecting informative genes from microarray expression data with biased sample distributions. / by Abu Hena Mustafa Kamal. / Thesis (M.S.C.S.)--Florida Atlantic University, 2009. / Includes bibliography. / Electronic reproduction. Boca Raton, Fla., 2009. Mode of access: World Wide Web.

The identification of drug treatments that are useful in diverse therapeutic settings is a significant driving force in biomedical research [Macarron et al., 2011], [Poureetezadi et al., 2014], [Lamb, 2007]. Typical means for measuring the efficacy of a drug for a given clinical application include protein-protein interactions, cell death, mitochondrial respiration and cell growth as well as broader measurements of absorption, distribution, metabolism, excretion and toxicity (ADMET), specifically related the the drug or drugs being tested [Szakcs et al., 2008]. A wide array of methods are routinely employed to perform these screens, from ligand binding assays [Wagner et al., 2016] to high-throughput proteomics [Verheul, 2014]. One method that is currently underutilized in small-molecule drug screens and drug discovery is high-throughput transcriptome sequencing, such as RNA-Seq. Although RNA-Seq is routinely used to profile patterns of genetic changes following perturbations such as drug treatment [Young et al., 2014], it has not, to my knowledge, yet been used as the primary readout of a drug screen.

Alzheimer’s disease (AD) affects tens of millions of people worldwide. Its cause is unknown, with no cure, and disease-modifying treatment options have only recently become available. Emerging research has made a strong case for the involvement of immune cells, such as microglia and T cells, in modulating AD pathology. Newer technologies in RNA-sequencing have detailed specific phenotypic changes to microglia and T cells over the course of neurodegenerative disease. Some researchers have also used whole-genome sequencing to correlate genetic variants with changes in gene expression. However, no studies thus far have conducted this type of genotype-phenotype correlation in immune cells from aged individuals or AD patients. We have collected gene expression data from four sorted T cell subtypes in peripheral blood samples from 96 subjects in ROSMAP, a cohort of AD patients and age-matched controls. 78 of these subjects also have whole-genome sequencing data, which we used to detect genetic variants associated with changes in T cell gene expression. These are known as expression quantitative trait loci (eQTL). We found genes related to T cell cytotoxicity and immunosenescence in gene co-expression modules, among the eQTL, and in correlation with AD neuropathological traits or risk variants for several disease traits. We extended our findings related to disease association by calculating polygenic risk scores (PRSs) in our cohort from whole-genome sequencing data for 19 traits related to immune function and disease, including AD. Genes associated with the PRS for one or more disease traits often were in biological pathways related to downstream cytokine signaling, regulation of T cell receptor signaling, and T cell migration and trafficking. Overall, our findings indicate that the use of aged and AD patients in T cell genotype-phenotype correlation studies highlights genetic variants and differentially expressed genes that are not seen in studies using young, healthy individuals.

Transcriptome studies disentangle functional mechanisms of gene expression regulation and may lend key insights into disease mechanisms. However, the cost of RNA-sequencing and types of tissues currently assayed pose major limitations to study expansion and disease-relevant discovery. This thesis develops methods for sampling noninvasive biospecimens for transcriptome studies, investigating their technical and biological characteristics, and assessing the feasibility of using noninvasive samples in transcriptomic and clinical applications. Chapter 1 explores the technical and biological features of four potential noninvasive sample types (buccal swabs, hair follicles, saliva, and urine cell pellets) in a pilot study of 19 individuals whereby four separate collections of each tissue were performed (i.e. 76 samples/tissue, 304 samples in total). From this data, consistency of library preparation, cell type content, replication of GTEx cis-eQTLs, and disease applications were assessed. In all, hair follicles and urine cell pellets were found to be most promising for future applications. Chapter 2 investigates the scaling potential of noninvasive sampling in SPIROMICS, a COPD clinical cohort. To do so, 140 hair follicle and 110 buccal swab samples were collected from seven different clinical sites. Consistency of sample quality was observed to be high for hair follicles, and hair cell type abundance estimates were consistent within SPIROMICS and compared to the 19 subject pilot study. Mapping of cis-eQTLs in hair revealed 339 associations not identified in any prior study. These cis-eQTLs show higher replication in GTEx tissues that share cell types with hair follicles, indicating hair follicles may indeed capture gene expression regulatory mechanisms found in more invasive tissue types of the body. This thesis suggests future use of noninvasive sampling will facilitate discovery by increasing sample sizes in more diverse populations and in tissues with greater cell type diversity and biological relatedness to disease mechanisms. Moreover, the nature of noninvasive sampling enables complex, longitudinal study designs with greater ability to capture context-dependent mechanisms of genetic regulation not currently able to be interrogated.