A New Gene Selection Technique Using Feature Selection Methodology

The DNA Microarray technology can measure the expression levels of thousands of genes simultaneously, and produces huge volumes of gene expression data. Such gene data include complex variations among expression levels of genes in the various classes of samples, which allows for classifying and clustering the samples based on only a small subset of genes. We aim to identify those genes that demonstrate high capabilities of discrimination between the classes of samples (e.g. the normal vs disease tissue samples). We present a new technique for gene selection and extraction using various feature selection techniques. Our method is based on computing thresholds and discriminating capabilities of each gene, and classifying the data according to only those genes that have highest discriminating capabilities. The method extracts very small subsets of informative genes that can improve the classification accuracy. We applied the method on four different common gene expression datasets used mainly for this purpose. The method produces encouraging and competitive results of classification performance compared with recent similar techniques.