Cancer-related Informative Gene Selection Based On Neighborhood Rough Set

Cancer seriously impact on human health and the quality of life.Thus,to explore cancer causative pathology,diagnosis and the development of effective new treatment for cancer has become the focus of medical experts and computer scientists.Due to the human has not really understand the pathogenic mechanisms of cancer yet,there is a big blindness of cancer treatment.The traditional method of cancer research has been unable to adapt to the current needs of the large-scale analysis of cancer research.The emergence of gene chip technology provides a strong database for the pathogenesis of cancer research and clinical diagnostic methods at the molecular level.In this paper,we have studied the selection method of the informative cancer genes using attribute reduction algorithm based on neighborhood rough set.The main innovation in this paper includes two aspects.First,this paper proposes a cancer-related gene selection method based on attribute reduction algorithm,which is based on the neighborhood rough set.It initially adopts the gene filtering method to tease out the differentially-expressed gene and then applies the attribute reduction algorithm to futher select the important cancer-related gene by searching out the best informative gene subset.The experimental results on the actual cancer gene expression datasets show that the constructed ensemble classifier can get stable forecast classific ation performance.By analyzing relevant biomedical literature,we verify the relevance of the selected genes and cancer,which proves the superiority of our method in finding important cancer-related genes.Second,it puts forward a method to construct ensemble classifier based on the attribute reduction algorithm.It firstly constructs the member classifier characterized with the selected gene subset by using the attribute reduction algorithm based on the neighborhood rough set,and then it builds up the ensemble classifier at the basis of the constructed member classifier.Regard to the principle that giving different value can select different gene subset,the varieties of extraordinary member classifier to assure their diversity are structured.And finally,the member classifiers are formed to an ensemble classifier by using the majority voting strategy,and we put it into the practice of the actual tumor gene expression data set,which can effectivly avoid the information loss of the discretization of the gene expression value.