Submit a Revision

*your email :**
page name* : such as "Introduction", "Conclusion"..etc
<h2> Comparative analysis of microarray normalization procedures: effects on reverse engineering gene networks</h2> <em><strong> Wei Keat Lim <sup>1</sup><sup>,2</sup>, Kai Wang <sup>1</sup><sup>,2</sup>, Celine Lefebvre <sup>2</sup> and Andrea Califano <sup>1</sup><sup>,2</sup><sup>,*</sup></strong></em> <p> <em> <sup>1</sup>Department of Biomedical Informatics, Columbia University, 622 West 168th Street, Vanderbilt Clinic 5th Floor and <sup>2</sup>Center for Computational Biology and Bioinformatics, Columbia University, 1130 Saint Nicholas Avenue, New York, NY 10032, USA </em> </p> <p> <em>An open access article: Bioinformatics 2007 23(13):i282-i288.</em> </p> <p> </p> <p> <strong>Abstract </strong><br> </p> <p> <strong>Motivation:</strong> An increasingly common application of gene expression<sup></sup>profile data is the reverse engineering of cellular networks.<sup></sup>However, common procedures to normalize expression profiles<sup></sup>generated using the Affymetrix GeneChips technology were originally<sup></sup>developed for a rather different purpose, namely the accurate<sup></sup>measure of differential gene expression between two or more<sup></sup>phenotypes. As a result, current evaluation strategies lack<sup></sup>comprehensive metrics to assess the suitability of available<sup></sup>normalization procedures for reverse engineering and, in general,<sup></sup>for measuring correlation between the expression profiles of<sup></sup>a gene pair.<sup></sup> </p> <strong>Results:</strong> We benchmark four commonly used normalization procedures<sup></sup>(MAS5, RMA, GCRMA and Li-Wong) in the context of established<sup></sup>algorithms for the reverse engineering of protein–protein<sup></sup>and protein–DNA interactions. Replicate sample, randomized<sup></sup>and human B-cell data sets are used as an input. Surprisingly,<sup></sup>our study suggests that MAS5 provides the most faithful cellular<sup></sup>network reconstruction. Furthermore, we identify a crucial step<sup></sup>in GCRMA responsible for introducing severe artifacts in the<sup></sup>data leading to a systematic overestimate of pairwise correlation.<sup></sup>This has key implications not only for reverse engineering but<sup></sup>also for other methods, such as hierarchical clustering, relying<sup></sup>on accurate measurements of pairwise expression profile correlation.<sup></sup>We propose an alternative implementation to eliminate such side<sup></sup>effect. <p> </p>
Confirmation code: Enter the code exactly as it appears. All letters are case insensitive, there is no zero.