Efficient gene selection with rough sets from gene expression data

Abstract

The main challenge of gene selection from gene expression dataset is to reduce the redundant genes without affecting discernibility between objects. A pipelined approach combining feature ranking together with rough sets attribute reduction for gene selection is proposed. Feature ranking is used to narrow down the gene space as the first step, top ranked genes are selected; the minimal reduct is induced by rough sets to eliminate the redundant attributes. An exploration of this approach on Leukemia gene expression data is conducted and good results are obtained with no preprocessing to the data. The experiment results show that this approach is successful for selecting high discriminative genes for cancer classification task. © 2008 Springer-Verlag Berlin Heidelberg.