Microarray Data Analysis Pipeline

Scanned images were first inspected for quality control (QC) using a variety of built-in QC tools from the Bioconductor package [10 ] of R, the open source environment for statistical analysis. QC consisted of visual examination of probe array images, scatter plots from replicates, hierarchical clustering of array hybridizations, RNA degradation plots and MvA plots. Feature intensity values from scanned arrays were normalised and reduced to expression summaries using the GC Robust Multiarray Algorithm (GCRMA) [11 ,12 (link)] implemented as a function in the Bioconductor GCRMA library [13 (link)]. Detection calls indicating the presence or absence of signal from each probe set were obtained by processing the raw data with the Microarray Analysis Suite 5.0 (MAS5). To obtain a consensus detection call across replicate hybridizations, a probe set was considered to be present if it received a P (present) detection call from all replicates or n-1 replicates with an M (marginal) call from the remaining replicate. A (absent) detection calls were determined in the same way.
For further analysis, probesets 3' locations were obtained by downloading the MOE430a probe tab files made available by the Affymetrix online support [39]. A probeset location was considered equal to the 3' distance of the probe that is most distal from the 3' end of the corresponding target within the set.
To test for differential expression, two statistical methods were used. The bayesian adjusted t-statistics from the linear models for Micoarray data (limma) package [14 ,15 ] and the Z-scores method as described by Quackenbush [16 ]. With both methods, a multiple testing correction based on the false discovery rate (FDR) was performed.