Research about feature genes selection for cancer type identification based on gene expression profiles

The identification and classification of different cancer type and feature gene subset selection are of great importance in cancer diagnosis and have recently received a great deal of attention in the field of bioinformatics. On the basis of comparing cancer with normal samples by SVM and verifying the disease group and normal group can be classified by the feature gene vectors, we selected the feature gene module of different cancer types in the training set with improved Relief algorithm, then put the feature gene module to the test set including 4 kinds of cancer samples. The results of series experiments in different conditions proved that the identification accuracy of selected feature genes can reach more than 95%. The SVM and improved Relief algorithm show excellent performance of selecting feature genes to identify and classify cancer types.