A hybrid method for gene selection in microarray datasets

Due to the steep price of a microarray experiment, a microarray dataset usually contains a few experimental samples. While the number of experimental samples is small, the number of genes in an experimental sample is quite large. The fact that only a few of the large amount of genes are relevant to a diagnosis poses a challenge to the application of the microarray technology. This paper presents a hybrid method to select important genes from a microarray dataset. The proposed method comprises three steps. In the first step, the information gains of genes in the dataset are calculated, and the genes with small information gains are eliminated from the dataset. In the second step, the remaining genes are clustered based on their pairwise correlation coefficients, and a representative gene is selected for each cluster to form the preliminary selected gene set. Finally, a genetic algorithm is used, in the third step, to further select genes from the preliminary gene set to generate the final selected gene set. The experiment result shows that the proposed method is better than the existing methods in terms of the classification accuracy and the number of selected genes.