Systematic order-dependent effect in expression values, variance, detection calls and differential expression in Affymetrix GeneChips registered

MOTIVATION: Affymetrix GeneChips registered are common 3' profiling platforms for quantifying gene expression. Using publicly available datasets of expression profiles from human and mouse experiments, we sought to characterize features of GeneChip registered data to better compare and evaluate analyses for differential expression, regulation and clustering. We uncovered an unexpected order dependence in expression data that holds across a variety of chips in both human and mouse data. RESULTS: Order dependence among GeneChips registered affected relative expression measures pre-processed and normalized with the Affymetrix MAS5.0 algorithm and the robust multi-array average summarization method. The effect strongly influenced detection calls and tests for differential expression and can potentially significantly bias experimental results based on GeneChip registered profiling. Supplementary information: Supplementary Material, including links to files of ordered transcripts and supporting analyses, is available at the authors' websites, at http://www-math.cudenver.edu/~kbjork/research/, and http://www-math.cudenver.edu/~kk/research/.