Optimal gene selection for cancer classification with partial correlation and k-nearest neighbor classifier

High density DNA microarrays are widely used in cancer research, monitoring thousands of genes at once. Due to small sample size and the large amount of genes in micrarray experiments, selection of significant genes via expression patterns is an important matter in cancer classification. Many gene selection methods have been investigated, but it is hard to find out the perfect one. In this paper we propose a new gene selection method based on partial correlation in regression analysis to find the informative genes to predict cancer. The genes selected by this method tend to have information about the cancer that is not overlapped by the genes selected previously. We have measured the sensitivity, specificity, and recognition rate of the selected genes with k-nearest neighbor classifier for colon cancer dataset. In most of the cases, the proposed method has produced better results than the gene selection methods based on correlation coefficients, showing high accuracy of 90.3% for colon cancer dataset.