A Probabilistic mechanism based on clustering analysis and distance measure for subset gene selection

Many subset gene selection methods for microarray data employ classification tools to evaluate the discernability of a gene subset on a specific disease, and this evaluation process generally has a high computational complexity. In this study, we propose a probabilistic mechanism supported by a density-based clustering method and a distance measure to perform individual and group gene replacement for gene selection. Analysts can choose proper values for the parameters of the probabilistic mechanism to set the computational complexity for gene selection. The discernability of a gene subset on classification is evaluated by the distance measure to avoid the language bias that can be introduced by classification tools. Our experimental results on six microarray data sets show that the probabilistic mechanism can effectively and efficiently filter a gene subset with a high discernability on cancer diagnosis.