Hiseq rna seqv2 rsem

Three publically available datasets were used in this study. These were chosen for several reasons: DC (lung tumors, both genders), GSE2034/Breast (same tumor type as data used in Venet et al. [2 (link)], single gender), and PRAD (prostate adenocarcinomas, single gender). The rationale for selecting these three datasets is the following: (1) the TvsN-100 signature is expected to perform well on all three datasets, (2) gender signature is expected to perform well on the DC dataset but not on Breast and PRAD datasets, (3) GSE2034/Breast was the same tumor type used by Venet et al. [2 (link)], and finally (4) they represent three different tumor types. The first is Director's Challenge (DC) dataset, which consists of 442 lung tumor samples [23 (link)] arrayed on the Affymetrix Human Genome U133A Array. This dataset was batch corrected using COMBAT [24 (link)], since it shows a clear dependence on the institution where the samples were run. The second dataset is from 286 Breast (Breast) samples (GSE2034) arrayed on the Affymetrix Human Genome U133A Array. Both of these datasets were normalized using IRON [25 (link)]. The third dataset is from The Cancer Genome Atlas (TCGA) and contains 297 primary prostate adenocarcinoma (PRAD). The level 3 Illumina HiSeq RNAseqV2 RSEM gene-level normalized mRNA expression data was downloaded from the TCGA data portal in December of 2014.