Mas 5.0 algorithm

We downloaded the TCGA RNA-Seq data (HTSeq Count) of EC using TCGAbiolink [25 (link)]. The TCGA MC3 files were retrieved from The Genomic Data Commons (GDC) portal to analyze the mutation profile of EC patients. The cBioportal was used to obtain the segmented copy number variation datasets [26 (link)]. We considered TCGA-Clinical Data Resource (CDR) to retrieve the corresponding clinical annotation for EC patients [27 (link)]. For validation, the microarray series dataset (GSE17025, Affymetrix Human Genome U133 Plus 2.0 Array), consisting of 91 EC samples, was downloaded from the NCBI GEO database using the GEOquery package [28 (link)]. We employed the human genome-scale metabolic model (HMR2.0) to study cancer metabolism. The model consists of 8181 reactions, 6006 metabolites, and 3765 genes, which describe the standard metabolic processes of a human cell.
The TCGA RNA-Seq dataset consists of 565 samples, of which 542 are primary tumor samples (sample type code—01) and 23 are tumor-matched normal samples (sample type code—11). We used variance-stabilizing transformation (VST) to normalize the RNA-Seq raw count data. The validation microarray dataset comprises 79 endometrioid and 12 papillary serous samples with various grades. We normalized the microarray data using Affymetrix’s MAS5.0 algorithm and log₂ transformation.

Affymetrix expression data for ColXa1 and Col11a1 genes in patient samples were produced from publicly available data sets of Moffitt Cancer Center patients. The CEL files for the tumor samples were downloaded from the Gene Expression Omnibus (GEO) database (http://www.ncbi.nlm.nih.gov/geo/), data series GSE2109. Normal tissue data were from the GEO data series GSE7307, Human Body Index. The CEL files were processed and analyzed using the MAS 5.0 algorithm (Affymetrix) and screened through a rigorous quality control panel to remove samples with a low percentage of probe sets called present by the MAS 5 algorithm, indicating problems with the amplification process or poor sample quality; high scaling factors, indicating poor transcript abundance during hybridization; and poor 30/50 ratios, indicating RNA degradation either before or during processing. The remaining samples were normalized to the trimmed average of 500 in the MAS 5 algorithm before comparison of the expression values across tumors and normal samples.

The gene expression analysis shown for candidate genes in medulloblastoma, other tumours and normal tissues was compiled from multiple gene expression profiling studies [5 (link),39 (link)–47 (link)] (Kool et al. unpublished data). All samples were analysed using the Affymetrix GeneChip Human Genome U133 Plus 2.0 arrays. The MAS5.0 algorithm of the GCOS program (Affymetrix Inc.) was used to normalize the expression data. Data were analysed and statistically verified using the R2 software platform for analysis and graphic visualization of microarray data (see http://r2.amc.nl).