DNA microarrays simultaneously gauge the expression of thousands of genes in clinical samples. In this paper, we focus on cancer studies, where gene expression technologies have been applied extensively (Alizadeh et al [1]; Khan et al [2]; Dhanasekaran et al [3]). Obtaining large-scale gene expression profiles of tumors should theoretically allow for the identification of subsets of genes that function as prognostic disease markers or biologic predictors of therapeutic response. Because the data are highly multivariate and complex, it is important to develop automated statistical methods to detect systematic signals in gene expression patterns.
