The application of the permutation test on genome wide expression analysis

• Main Author: Chan, Timothy
• Language: English
• Published: 2010
• Online Access: http://hdl.handle.net/2429/17660

We are now in a new era. The recent completion of the entire sequence of the human genome and high-throughput gene expression technologies has transformed the era of molecular biology to the era of genomics. Already, such technologies are showing great promise in disease classification and gene targets. However, like any new exciting technology, great promise and anticipation can lead to wasted resources and false hope. It is critical that we recognize the experimental limitations of these new technologies and most importantly, hidden problems must be addressed. The primary goal of a high-throughput gene expression experiment is to identify genes of interest that are differentially expressed between two sample groups. This thesis addresses two key issues that have hindered high-throughput gene expression technologies. The first is the sample size issue. Small sample sizes affect statistical confidence and are much more sensitive to outliers. Thus, we show that by using a nonparametric statistical test known as the permutation test, we can achieve higher accuracy than conventional parametric statistical tests such as the t-test. The second issue we address is the use of housekeeping genes for normalization of mRNA levels. It is well known that many biological experiments require a set of reference genes that are highly expressed and constant from sample to sample. The choice of reference genes is critical as the wrong choice can have dire effects on subsequent analyses. To address this issue, we developed a methodology based on SAGE, which is a genome wide expression technology that does not require normalization. Our results suggest that reference genes chosen by our methodology are more appropriate for mRNA normalization than the standard set of housekeeping genes. Furthermore, our results suggest that reference genes are more effective if chosen in a tissue-specific manner. === Science, Faculty of === Computer Science, Department of === Graduate