A novel approach for dissimilar gene selection and multi-class classification of neuromuscular disorders: combining median matrix and radial basis function based support vector machine

Abstract

Accurate prediction of the kind of neuromuscular disorder (NMD) is fundamental for choosing optimal treatment for patients. Here we intend to select the compact subsets of dissimilar and discriminating genes from thousands of genes that can successfully classify various kinds of NMDs. We propose a new integrated model for gene selection and multi-class classification of NMDs. The gene expression matrix is processed to create a median matrix for the selection of compact and different subsets of genes for every class. The classification algorithms use the combination of selected genes for prediction of the kind of NMD samples. The accuracy and effectiveness of the proposed model are exhibited through analysis of publicly available microarray dataset of 13 NMDs. The integration of the proposed method of gene selection with RBF SVM classification algorithm has outperformed in most of the cases. The results confirm the ability of the proposed model for identifying the subsets of most discriminating, non-redundant and dissimilar genes which helps the classifier to give a high classification performance.