A Gene Selection Software Package for Cancer Classification

Selecting a small number of relevant genes for accurate classification of samples is essential for the development of diagnostic tests, which have been the subject of considerable research in the past few years. However, many researches have still been trying to improve the algorithms to obtain better results. Here we present a novel implementation of recursive feature elimination method (nRFE) for gene selection and classification of microarray data. Our algorithm was evaluated over the NCI60 benchmark datasets, with an accuracy of 96.6% in 10-fold cross-validation, respectively. Furthermore, the nRFE outperformed recently published algorithms when applied to another two multi-cancer data sets. Computational evidence indicated that nRFE can avoid overfitting effectively. The combination of high accuracy and small numbers of genes should make nRFE a powerful tool for gene selection from gene expression data.