Automated diagnosis and prognosis of tumors of the central nervous system (CNS) offer overwhelming challenges because of heterogeneous phenotype and genotype behavior of tumor cells (Yang et al. 2003, Pomeroy et al. 2002). Unambiguous characterization of these tumors is essential for accurate prognosis and therapy. Although the present imaging techniques help to explore the anatomical features of brain tumors, they do not provide an effective means of early detection. Currently, the histological examination of brain tumors is widely used for an accurate diagnosis; however, the tumor classification and grading based on histological appearance does not always guarantee absolute accuracy (Yang et al., 2003, Pomeroy et al., 2002). In many cases, it may not be sufficient to detect the detailed changes in the molecular level using a histological examination (Yang et al. 2003) since such examination may not allow accurate prediction of therapeutic responses or prognosis. If the biopsy sample is too small, the problems are aggravated further. Toward achieving a more reliable diagnosis and prognosis of brain tumors, gene expression measures from microarrays are the center of attention to many researchers who are working on tumor prediction schemes. Our proposed tumor prediction scheme is discussed in two chapters in this volume. In part I (this chapter), we use an analysis of variance (ANOVA) model for characterizing the Affymetrix gene expression data from CNS tumor samples (Pomeroy et al. 2002) while in part II we discuss the prediction of tumor classes based on marker genes selected using the techniques developed in this chapter. In this chapter, we estimate the tumor-specific gene expression measures based on the ANOVA model and exploit them to locate the significantly differentially expressed marker genes among different types of tumor samples. We also provide a novel visualization method to validate the marker gene selection process.
Numerous statistical methods have evolved that are focused on the problem of finding the marker genes that are differentially expressed among tumor samples (Pomeroy et al., 2002, Islam et al., 2005, Dettling et al., 2002, Boom et al., 2003, Park et al., 2001). For example, Pomeroy et al. (2002) uses student t-test to identify such genes in embryonal CNS tumor samples. Because of the non-normality of gene expression measurements, several investigators have adopted the use of nonparametric methods, such as the Wilcoxon Sum Rank Test (Wilcoxon, 1945) as a robust alternative to the parametric procedures. In this chapter, we investigate a Wilcoxon-type approach and adapt the resulting procedures for locating marker genes.
Typically, statistical procedures for microarray data analysis involve performing gene specific tests. Since the number of genes under consideration is usually large, it is common practice to control the potentially large number of false-positive conclusions and family-wise error rates (the probability of at least one false positive statement) through the use of P-value adjustments. Pollard et al. (2003) and Van der Laan et al. (2004a, 2004b, 2005c) proposed methods to control family-wise error rates based on the bootstrap resampling technique of Westfall & Young (1993). Benjamini & Hochberg (1995), Efron et al. (2001) and Storey et al. (2002, 2003a, 2003b, 2004) introduced various techniques for controlling the false discovery rate (FDR), which is defined as the expected rate of falsely rejecting the null hypotheses of no differential gene expression. These adjustment techniques have gained prominence in statistical research relating to microarray data analysis. Here, we use FDR control because it is less conservative than family-wise error rates for adjusting the observed P-values for false discovery. In addition, we propose a novel marker gene visualization technique to explore appropriate cutoff selection in the marker gene selection process.
Key Terms in this Chapter
Kruskal-Wallis Test: A nonparametric mean test which can be applied if the number of sample groups is more than two,unlike the Wilcoxon Rank Sum Test.
Wilcoxon Rank Sum Test: A nonparametric alternative to the two sample t-test which is based on the order in which the observations from the two samples fall.
DNA Microarray: A collection of microscopic DNA spots, commonly representing single genes, arrayed on a solid surface by covalent attachment to chemically suitable matrices.
Parallel Coordinates: A data visualization scheme that exploits 2D pattern recognition capabilities of humans. In this plot, the axes are equally spaced and are arranged parallel to one another rather than being arranged mutually perpendicular as in the Cartesian scenario.
q-values: A means to measure the proportion of FDR when any particular test is called significant.
Histologic Examination: The examination of tissue specimens under a microscope.
False Discovery Rate (FDR): Controls the expected proportion of false positives instead of controlling the chance of any false positives. An FDR threshold is determined from the observed p-value distribution from multiple single hypothesis tests.