A tuning parameter selector method for the gSVM-SCAD in determining significant genes and biological pathways

Abstract

A hybrid of support vector machines and a smoothly clipped absolute deviation with group-specific penalty terms (gSVM-SCAD) is a penalized classifier that has been used to identify and select significant pathways in pathway-based microarray analysis. Despite its advantages in identifying significant pathways, the gSVM-SCAD has some limitations, as it depends on the proper choice of tuning parameter. If the tuning parameter is too small, it can bring little sparsity and overfit to the classifier model, while if it is too large, it can make much sparsity to the classifier model and produce poor discriminating power. Therefore, it is important to choose an appropriate tuning parameter selector method for the gSVM-SCAD. The generalized cross validation (GACV) has been widely used as a tuning parameter selector method. Unfortunately, GACV has some limitations where it poorly performs when dealing with the low number of variables (in this paper referred as genes) and large sample sizes. This is because some pathways contain not more than 100 genes and even some pathways contain less than 10 genes. This scenario can lead to the poor performance of SCAD in selecting the informative genes and simultaneously identifying significant ones. In order to surmount the limitations of the gSVM-SCAD, we proposed to use the B-type generalized approximate cross validation (BGACV) as a tuning parameter selector method for gSVM-SCAD.