Hybrid PCA and LDA Analysis of Microarray Gene Expression Data

Abstract

Microarray technology offers a high throughput means to study expression networks and gene regulatory networks in cells. The intrinsic nature of high dimensionality and small sample size in microarray data calls for the development of effective computational methods. In this paper, we propose a novel hybrid dimension reduction technique for classification - hybrid PCA (principal component analysis) and LDA (linear discriminant analysis) analysis. This technique effectively solves the singular scatter matrix problem caused by small training samples and increases the effective dimension of the projected subspace. It offers more flexibility and a richer set of alternatives to LDA and PCA in the parametric space. In addition, generalization of hybrid analysis of other dimension reduction techniques is also proposed in this paper, such as multiple discriminant analysis (MDA) and biased discriminant analysis (BDA). Extensive experiments on the yeast cell cycle regulation data set show the superior performance of the hybrid analysis over the traditional methods such as SVM.