Study of Informative Gene Selection for Gene Expression Profiles

In the research community of gene expression microarray data analysis, selecting a small number of informative genes from thousands of genes is a hot research problem for accurate classification of diseases. In this paper, we present our gene selection algorithm SIGWeight. SIGWeight expands sigmoid kernel into its Maclaurin series, and calculate the weight of each feature(gene) from the classification hyperplane which is constructed by SVM with SIGMOID kernel, and then sort the features according to the weight of features. A subset of features is selected from top of the ranking list. We apply four widely used classifiers on the obtained datasets with only the selected genes to evaluate effectiveness of SIGWeight. Experiment results on Leukemia and DLBCL datasets show that SIGWeight is very encouraging, Furthermore, compared with information gain(IG), SIGWeight select less number of genes, At the same time, the genes selected led to the highest accuracy.