Selected Based On The Gene Expression Profiles Of Tumor Characteristic Gene Studies

With the successfully completion of human genome project and the further perfection of gene chip technology, tumor research has entered into a new era. The pathogenesis of cancer is complex with the various types, not easy to be treated. So the early diagnosis and the type determination of tumor could help patients save their lives. Now, the diagnosis and the classification of cancer is based on the observation of morphological above histopathological and through immune biochemical characteristics to identify types. However, it does have a big defect it is heavily reliant on the personal subjective experience of tumor tissue pathology to classify the tumor types. On the other hand, some pathology feature of tumor tissue is very similar without remarkable differences, which make the specificity of the cancer treatment become not efficient.DNA chip technology with a high throughput characteristics, also can measure multi-gene expression patterns of cells massively, and then the gene expression profiles can be obtained. The difference of gene expression which inside of tumor tissue cell reflects a higher specificity relying on the changing of cancer gene expression profile expression value; the similar tumor tissue could be discriminated. The research based on the gene expression profile could improve the quality of diagnosis and treatment.This paper presents a method of feature gene selection based on relative risk and feature weighting RR-FW. Gene expression profile has some characteristics, such as small sample, high dimension, high noise, The difference expression of mutated gene in tumor cells is more remarkable than normal tissue cells. RR (relative risk) can be used to mine the genes with the difference expression. The bigger relative risk value is with the more significant the difference of gene expression. Euclidean distance is used to measure the correlation between the feature vectors and calculate the k-nearest sample of the same category and the different category. The strength of the classification ability to identify types and subtypes of tumor by the feature genes is measured by the size of weights. Redundancy coefficients between of two genes are used to remove strong redundancy genes. Dimensions, lots of noise and redundant genes are fully reduced and avoided interference of the nonspecific error. Through training and testing of the multiple classifiers, comparing with other methods, the proposed is superior to other methods, and with minimal feature genes, classification accuracy can reach100%, the precision and generalization performance of the classifiers is improved with the low space computing complexity. This method is faster, better versatility, and the found feature genes have biological significance. The proposed approach is useful for cancer clinical diagnosis and determining treatment of drug molecules target, it also helps to reveal the pathogenesis and biological function of oncogenes.