Assessing the efficiency of dye-swap normalization to remove systematic bias from two-color microarray data

Microarrays are a powerful tool in functional genomics, what allow a simultaneous analysis of the expression level of thousands of genes under different conditions. In order to compare measurements within and across arrays and to correct non-biological variation masking meaningful information, normalization is an essential task prior to any further analysis. Among all the available normalization techniques, LOWESS has proved useful for the normalization of data generated from the two main microarray platforms (two-color arrays and affymetrix chips) due to its ability to remove intensity dependent effects. However, the use of this robust estimator to correct the data without taking account of biological characteristics is a concern often raised by microarray analysts. In addition, powerful software packages are needed to perform such computationally expensive normalization and several parameters need to be fixed in advance, resulting in differently corrected data sets for the different sets of parameters used. Reverse labeling designs are common setups in two-color microarray experiments if comparison between the co-hybridized samples is of interest. Using three different data sets, this paper assesses the effectiveness of dye-swap normalization, a method that makes use of the intrinsic information provided by this type of experimental design. The results show how dye-swap normalization corrects the bias introduced by the different properties of the dyes, removing intensity dependent effects. Furthermore, dye-swap normalization corrects the data accounting for gene-dye effects and the transformation of the data is justified on a biological basis. The results present dye-swap normalization as a valid alternative to normalize two-color microarray data. The paper also reviews the assumptions made and the formulas applied to correct the data using dye-swap normalization. In addition, a generalization of the dye swap normalization formula is implemented to normalize data generated from microarray experiments for which a large proportion of genes are expected to be differentially expressed.∗ The figures and results presented in this paper have been implemented using several Bioconductor packages from R, MATLAB (Mathworks Inc.) and ArrayNorm. A collection of files and supplementary material is available from http://www.sbi.uni-rostock.de