A multivariate algorithm for gene selection based on the nearest neighbor probability

Abstract

Experiments performed with DNA microarrays have very often the aim of retrieving a subset of genes involved in the discrimination between two physiological or pathological states (e.g. ill/healthy). Many methods have been proposed to solve this problem, among which the Signal to Noise ratio (S2N) [5] and SVM-RFE [6]. Recently, the complementary approach to RFE, called Recursive Feature Addition (RFA), has been successfully adopted. According to this approach, at each iteration the gene which maximizes a proper ranking function φ is selected, thus producing an ordering among the considered genes. In this paper an RFA method based on the nearest neighbor probability, named NN-RFA, is described and tested on some real world problems regarding the classification of human tissues. The results of such simulations show the ability of NN-RFA in retrieving a correct subset of genes for the problems at hand. © 2009 Springer Berlin Heidelberg.