A New Nonparametric Gene Selection Method for Classification of Microarray Data

摘要

Gene selection is a central step of gene expression data analysis. In this paper, a new nonparametric method, Gene Selection for Multiclass (GSM), is proposed, which selects genes based on the criterion of the large inter-class difference and the small intra-class difference. Using the default training and testing sets on two publicly available datasets, leukemia (two classes) and SRBCT(four classes), the proposed method has been evaluated and compared with three relative methods, F-test, SAM and cho. The experimental results show GSM is effective and robust to select differential expression genes.