Illumina BeadStudio version 3.0 (Illumina, San Diego, CA, USA) was used for all low-level preprocessing steps of the expression data, including background estimation and correction, normalization and probe set summary. After these low-level preprocessing steps, 9875 genes with high detection P-value (>0.1) or >10% missing values were filtered out and removed from the data set.25 The remaining missing signal intensities were estimated using the ‘k nearest neighbor' algorithm implemented in R BioConductor.26, 27 The resulting data set was subsequently log2 transformed. Finally, after all preprocessing steps, the raw data of 48 701 probe signal intensities were mapped and reduced to signal intensities corresponding to 15 439 unique genes.
