Publication Year:
Usage 274
Downloads 274
Repository URL:
Pitt D-Scholarship Id:
Changqing Ma
thesis / dissertation description
Two major issues hinder the application of microarray based gene expression profiling in clinical laboratories as a diagnostic or prognostic tool. The first issue is the sheer volume and high-dimensionality of gene expression data from microarray experiments, which require advanced algorithms to extract meaningful gene expression patterns that correlate with biological impact. The second issue is the substantial amount of variation in microarray gene expression data, which impairs the performance of analysis method and makes sharing or integrating microarray data very difficult. Variations can be introduced by all possible sources including the DNA microarray technology itself and the experimental procedures. Many of these variations have not been characterized, measured, or linked to the sources. In the first part of this dissertation, a decision tree learning method was demonstrated to perform as well as more popularly accepted classification methods in partitioning cancer samples with microarray data. More importantly, results demonstrate that variation introduced into microarray data by tissue sampling and tissue handling compromised the performance of classification methods. In the second part of this dissertation, variations introduced by the T7 based in vitro transcription labeling methods were investigated in detail. Results demonstrated that individual amplification methods significantly biased gene expression data even though the methods compared in this study were all derivatives of the T7 RNA polymerase based in vitro transcription labeling approach. Variations observed can be partially explained by the number of biotinylated nucleotides used for labeling and the incubation time of the in vitro transcription experiments. These variations can generate discordant gene expression results even using the same RNA samples and cannot be corrected by post experiment analysis including advanced normalization techniques. Studies in this dissertation stress the concept that experimental and analytical methods must work together. This dissertation also emphasizes the importance of standardizing the DNA microarray technology and experimental procedures in order to optimize gene expression analysis and create quality standards compatible with the clinical application of this technology. These findings should be taken into account especially when comparing data from different platforms, and in standardizing protocols for clinical applications in pathology.