The summary probe-level data delivered by the Illumina scanner (mean and SD computed over all beads for a particular probe) was loaded in Beadstudio. The pre-processing done by the Illumina software, at the level of the scanner and by Beadstudio included: correction for local background effects, removal of outlier beads, computation of average bead signal and SD for each probe and gene, calculation of detection P-values using negative controls present on the array, quantile normalization across arrays, check of outlier samples using a clustering algorithm, check of positive controls. Analyses were carried out on the mean level for all probes in each gene. To stabilise variance across expression levels, we applied an arcsinh transformation to the expression data [55]. Compared to a log transformation, this transformation has the advantage not to discard negative expression values which can occur in Illumina data.
