Outcomes of gene association analysis of cancer microarray data are impacted by pre-processing algorithms

Abstract

Gene association analysis of cancer microarray data provides a wealth of information on gene expression patterns and cancer pathways to enhance the identification of potential biomarkers for cancer diagnosis, prognosis, and prediction of therapeutic responsiveness. However, achieving these biological/clinical objectives relies heavily on the functional capabilities and accuracy of the various analytical tools to mine these cancer microarray gene expression profiles. Many preprocessing algorithms exist for analyzing Affymetrix microarray gene expression data. Previous studies have evaluated these algorithms on their capabilities in accurately determining gene expression using a variety of spike-in as well as experimental data sets. However, variations in detecting differentially expressed genes between these different pre-processing algorithms on a single cancer dataset have not been done in a systems-level evaluation. In this study, we assessed the comparability and the level of variation between PLIER, GCRMA, RMA and MAS5 for their capability to detect differentially expressed genes.