Dimensionality reduction and discrimination

With microarrays and other high-throughput genomic profiling technologies that allow for the simultaneous analysis of expression levels of thousands or even tens of thousands of genes, we are able to expand our ability to characterize and understand disease processes at the molecular level and the heterogeneity surrounding them. As advances in biotechnology continue to support this remarkable expansion, however, the need for extracting and synthesizing information from the volumes of expression data has created an equally challenging research area in the development of corresponding statistical methods for their analysis. Within the context of biomedical and clinical research, a common objective of microarray data analysis is the selection of genes, from among the thousands profiled, that characterize groups of similar phenotype based on a small number of samples.