Algorithm for gene selection from DNA-microarray data for disease classification

In recent years, microarray technology has advanced to such a sophistication that, it is possible to obtain gene-expression level of several thousand genes in a single experiment. Simultaneous measurements of tens of thousands of mRNAs can be performed, in which gene expressions of two samples are compared. Depending on the source of the two compared samples, important investigations, like disease progress, diagnosis, drug response, etc., can be done by analyzing DNA microarray data. When one sample source is a healthy cell, and the other a cancerous one, it is possible to identify changes in particular gene expression with the progress of the disease. The aim of this work is to identify a few number of genes, which as a set of features, could clearly classify the target disease. The problem is defined as an optimization problem, where the target is to find minimum number of genes whose expression data could classify the disease type with minimum classification error. As we view the genes as features, the whole microarray data is of enormously high dimensional, where expression value of most of the genes are irrelevant to the targeted investigation. Moreover, the number of samples are in tens to a maximum of around hundred. Under such situation, identifying and eliminating irrelevant genes is of utmost importance. In this paper, we present a two stage reduction. In Stage 1, the number of genes are reduced from thousands to around hundred. We propose a new algorithm for Stage 1 reduction phase. In Stage 2, the number of selected genes are only a few. We proposed a way to achieve that optimization, without actual experiments.

I. Introduction

DNA microarray technology facilitates the measurement of the gene expression levels of several thousand genes in a single experiment [21] [22]. It is important to know the procedure for obtaining gene expression data, to understand its analysis and various applications. The procedure is briefly described here.