Cross phenotype normalization of microarray data

Normalization is a prerequisite for almost all follow-up steps in microarray data analysis. Accurate normalization across different experiments and phenotypes assures a common base for comparative yet quantitative studies using gene expression data. In this paper, we report a comparison study of four normalization approaches, namely, linear regression (LR), Loess regression, invariant ranking (IR) and iterative nonlinear regression (INR), for gene expression data normalization. Among these four methods, LR and Loess regression methods use all available genes to estimate either a linear or nonlinear normalization function, while IR and INR methods feature some iterative processes to identify invariantly expressed genes (IEGs) for nonlinear normalization. We tested these normalization approaches on three real microarray data sets and evaluated their performances in terms of variance reduction and fold-change preservation. By comparison, we found that (1) LR method exhibited the worst performance in both variance reduction and fold-change preservation, and (2) INR method showed an improved performance in achieving low expression variance across replicates and excellent fold-change preservation for differentially expressed genes.