Methods for enhanced detection and analysis of differentially expressed genes using gene chip microarrays

Abstract

A method for enhanced detection and statistical analysis of differentially expressed genes in gene chip microarrays employs: (a) transformation of gene expression data into an expression data matrix (image data paradigm); (b) wavelet denoising of expression data matrix values to enhance their signal-to-noise ratio; and (c) singular value decomposition (SVD) of the wavelet-denoised expression data matrix to concentrate most of the gene expression signal in primary matrix eigenarrays to enhance the separation of true gene expression values from background noise. The transformation of gene chip data into an image data paradigm facilitates the use of powerful image data processing techniques, including a generalized logarithm (g-log) function to stabilize variance over intensity, and the WSVD combination of wavelet packet transform and denoising and SVD to clearly enhance separation of the truly changed genes from background noise. Detection performance can be assessed using a true false discovery rate (tFDR) computed for simulated gene expression data, and comparing it to estimated FDR (eFDR) rates based on permutations of the available data. Where a small number (N) of samples in a group is involved, a pair of specific WSVD algorithms are employed complementarily if N>5 and if N<6.