Sensitivity and Consistency of Affymetrix GeneChip Normalization Methods

Many algorithms that calculate expression values from raw probe level data in Affymetrix GeneChips use a set of predetermined arrays for normalization. In this paper we analyze the sensitivity and consistency of expression values to the set of samples used for normalization. Our fundamental assumption is that expression levels of individual genes should strongly correlate with themselves as we vary the size and composition of the normalization set. We call this property "consistency". We varied the size and composition of the normalization set for publicly available Affymetrix GeneChip Microarrays and found that some normalization methods were consistent while others were highly inconsistent. It is fundamentally important for researchers to be aware of the consistency of the normalization methods they use while conducting microarray data analysis.