Algorithm for selection of informative genes using gene expression data

Abstract

Informative gene selection from high dimensional gene expression data has appeared as an important area of research in agri-genomics. Different gene selection techniques have been developed in recent times based on relevancy and redundancy of genes with class and among the genes. Most popular techniques for informative gene selection are Maximum Relevancy and Minimum Redundancy (MRMR) and Support Vector Machine Recursive Feature Elimination (SVM-RFE). However, these methodology have some drawbacks. One of the major drawback is that it ignores the spurious relations between genes and trait under study. In this study, a methodology for informative gene selection has been developed, which takes care of this spurious relation by implementing the bootstrap technique along with SVM-RFE and MRMR. The performance of these gene selection techniques has been analysed through classification accuracy of the SVM model with linear kernel developed using selected informative genes as predictors. A comparative evaluation of the developed method was done against three well known existing techniques for gene selection viz. Boot-MRMR, SVM-RFE, MRMR. On the basis of various evaluation measures, it has been observed that the performance of the developed methodology is better as compared to above given techniques and select less number of more informative genes. Moreover, for proper implementation and dissemination of the developed methodology, a user friendly R software package named “IGST” has been developed by using state of the art technology.