Samples of purified total RNA from six human cancer cell lines were purchased from Life Technologies. One gene-expression profile per sample was generated using the Affymetrix GeneChip HG-U133 Plus 2.0 Array, the Illumina Human HT-12 v4 Expression BeadChip Array and mRNA-seq (Illumina Hi-Seq) by Expression Analysis, a genomics contract research organization. The L1000 samples were profiled in multiple replicates. Data were normalized within platform (level 3, see below for details). For each cell line, we selected the L1000 replicate with highest technical quality (by LISS goodness of fit, see below) for comparison with the other three platforms. We then performed ComBat batch correction to adjust for cross-platform differences (Johnson et al., 2007), and subjected the data to hierarchical clustering in the space of the 952 genes commonly measured by all four platforms. We observe that the data cluster by cell line and not by platform, suggesting that the cross-platform differences are smaller than the biological differences between cell lines.
