Joint Discriminatory Gene Selection for Molecular Classification of Cancer

Abstract

This chapter introduces gene selection approaches in microarray data analysis for two purposes: cancer classification and tissue heterogeneity correction and hence is divided into two respective parts. In the first part, we search for jointly discriminatory genes which are most responsible to classification of tissue samples for diagnosis. In the second part, we study tissue heterogeneity correction techniques, in which independent component analysis is applied to tissue samples with the expression levels of only selected genes, the genes which are functionally independent and/or jointly discriminatory; we also employ non-negative matrix factorization (NMF) to computationally decompose molecular signatures based on the fact that the expression values in microarray profiling are non-negative. Throughout the chapter, a real world gene expression profile data was used for experiments, which consists of 88 tissue samples of 2308 effective gene expressions obtained from 88 patients of 4 different neuroblastoma and non-hodgkin lymphoma cell tumors.