Reduct Generation and Classification of Gene Expression Data

Abstract

Identification of gene subsets responsible for discerning between available samples of gene microarray data is an important task in bioinformatics. Due to the large number of genes in samples, there is an exponentially large search space of solutions. The main challenge is to reduce or remove the redundant genes, without affecting discernibility between objects. Reducts, from rough set theory, correspond to a minimal subset of essential genes. We present an algorithm for generating reducts from gene microarray data. It proceeds by preprocessing gene expression data, discretization of real value attributes into categorical followed by positive region based approach for reduct generation. For comparison, different approaches for reduct generation have also been discussed. Results on benchmark gene expression datasets demonstrate more than 90% reduction of redundant genes