Chemometric strategies for normalisation of gene expression data obtained from cDNA microarrays

DNA microarray technology is emerging as an important method in Functional Genomics. Microarrays enable rapid, large scale gene expression analysis but the cumulative, systematic errors which occur during experimentation result in large variances in the results. Thus data pretreatment, or normalisation, has been recognised as a major challenge in the analysis of microarray data. In this study, chemometric strategies were applied to microarray data in order to develop an appropriate normalisation method. In total, 192 pretreatment methods were systematically designed based on using the empty spots as background. In order to find the best method, partial least squares (PLS-1) was used to classify microarray data into compound treated and control groups. The optimal technique -a robust normal method (RNM), was identified by itʹs correct classification rate (CCR) based on cross-validation. In the RNM, the microarray data were normalised using signals of the k nearest neighbour (kNN) empty spots to compensate for localised background variation. The results of our study showed that RNM greatly reduced the variances within slides and between slides, performing better than any other studied normalisation methods, increasing the CCR by 17% and greatly improving the effective reproducibility of microarray data. Thus, RNM provides an effective method for normalisation of gene expression data obtained from DNA microarrays.