Open Access Highly Accessed Methodology article

Iterative rank-order normalization of gene expression microarray data

Eric A Welsh*, Steven A Eschrich, Anders E Berglund and David A Fenstermacher

Author Affiliations

H. Lee Moffitt Cancer Center and Research Institute, University of South Florida, Tampa, FL, 33612, USA

For all author emails, please log on.

BMC Bioinformatics 2013, 14:153  doi:10.1186/1471-2105-14-153

Published: 7 May 2013

Additional files

Additional file 1: Figure S1:

Background-subtracted normalization. Scatterplots are of log10 background-subtracted probe intensities. Points are colored by density (red: high, blue: low). Background subtraction was performed prior to normalization, reflecting the behavior of normalization within the IRON (A) and RMA (B) pipelines. IRON normalization centers the highest density distribution along the diagonal (thick green line), while quantile normalization centers the region between the two density distributions along the diagonal. Although generally down-shifted in intensity, the same patterns are observed in the background-subtracted data as in non- background-subtracted examples.

Format: PNG Size: 179KB Download file

Open Data

Additional file 2: Figure S2:

IRON vs. fixed-rank pair-wise normalization. Scatterplots are of log10 non- background-subtracted probe intensities. Points are colored by density (red: high, blue: low). Iterative rank order normalization (B), with a gradually decreasing rank-difference cutoff, is more robust to symmetry violations than a fixed rank-difference cutoff of 0.5% (A), and better centers the distribution of highest density along the diagonal (thick green line) than dChip (C).

Format: PNG Size: 237KB Download file

Open Data