Model-based Variance-stabilizing Transformation for Illumina Mi-croarray Data

Simon M. Lin, Pan Du and Warren A. Kibbe
Robert H. Lurie Cancer Center, Northwestern University, Chicago, IL 60611


Motivation: Variance stabilization is critical for subsequent statisti-cal inference to identify differential genes from microarray data. Due to limited technique replicates for Affymetrix and cDNA arrays, achieving variance stabilization is difficult. Although Illumina microar-ray provides larger number of technical replicates (30 to 40 beads per gene, randomly distributed on each slide), this unique feature is not yet leveraged in the current log2 data transformation process. Results: We devised a variance-stabilizing transformation (VST) by taking the advantage of larger number of technical replicates avail-able on the Illumina microarray. A robust spline normalization (RSN) algorithm, which combines the features of quantile and loess nor-malization, is designed to normalize the variance-stabilized data. We have evaluated the algorithms with the Barnes (2005) titration data set. We suggest that the VST algorithm, when used in tandem with RSN and empirical Bayes correction of the linear model (LIMMA), can constitute an optimal pipeline to analyze Illumina microarray data, outperforming the current strategy of logarithmic transforma-tion followed by quantile normalization.  

Correspondence to: Simon Lin
Tel: (+1) 312 695 1331
Fax: (+1) 312 695 1347 Publication URL Link to the journal's website. TBA
Publication Citation: Lin SM, Du P and Kibbe WA, Model-based Variance-stabilizing Transformation for Illumina Mi-croarray Data, 2006 (submitted)

lumi package

lumi annotation packages