We sought to compare expression profiles generated using L1000 with those generated using Affymetrix and RNA-seq, the most widely employed platforms for gene expression profiling. In conjunction with the NIH's Genotype Tissue Expression (GTEx) project (http://commonfund.nih.gov/GTEx/index), we profiled 3,176 samples on L1000 and obtained from GTEx the RNA-seq (Illumina TrueSeq RNA sequencing) data for different aliquots of these same samples (DSGEO-RNA-seq and DSGEO-L1000). The data were quantile normalized independently by platform (level 3, see below) and then batch-corrected using the ComBat algorithm, an empirical Bayes-based method commonly used to remove batch effects across gene expression datasets (Johnson et al., 2007).
