Adria Caballe Mestres email@example.com
Four publicly available Affymetrix microarray datasets were downloaded from the NCBI GEO repository . These datasets included gene expression and clinical information from a total of 1,072 colorectal cancer (CRC) patients. GSE14333  is a pool of 290 patients with CRC treated at 2 different hospitals: the Peter MacCallum Cancer Center (Australia) and the H. Lee Moffitt Cancer Center (United States); GSE33113  contains samples from 90 AJCC stage II patients collected at the Academic Medical Center in Amsterdam (the Netherlands); GSE39582  includes data from 566 CRC patients that form part of the Cartes d’Identite des Tumeurs (CIT) program, from the French ligue nationale contre le cancer; finally, GSE37892  includes expression and clinical information from 130 stage ii and iii CRC patients collected at five different hospitals from France (Marseille la timone, Nice lacassagne, Marseille institut paolicalmettes, Paris lariboisiare, Nancy brabois and Paris saintantoine).
Processing of microarray samples was carried out separately for each dataset using packages affy  and affyplm  from Bioconductor . Raw cel files were normalized using RMA background correction and summarization . Standard quality controls were performed in order to identify abnormal samples. Technical information concerning sample processing and hybridization was retrieved from the original CEL files: scanning dates were collected in order to define scan batches in each dataset separately; technical metrics PM MED, PM IQR, RMA IQR and RNA DEG described in  were computed and recorded as additional features for each sample. Probeset annotation was performed using the information available in Affymetrix. No sample was excluded due to quality issues.
Microarray datasets were corrected separately by metrics PM IQR, RMA IQR and RNA DEG. For doing so, a linear model was fitted separately for each probeset that included these metrics as the only explanatory variables, and coefficients of such models were used to correct the expression values a-priori. Next, a second linear model was fitted to each probeset and dataset separately, in order to correct by potential technical effects captured by sample’s center of origin and batch (scanning day). This correction was carried out using a mixed-effect model in which gender, age at diagnosis, stage, tumour location and Microsatellite Instability (MSI) status were also included as covariates, when available. Scanning day was modeled as a random effect in these models, while center was included as a fixed (GSE14333) or a random effect (GSE39582 and GSE37892) depending on the number of centers involved and on the sample size in each of them. Expression intensities were summarized at the gene level (entrez) by the first principal component of the probesets mapping to the same gene. This component was centered and scaled to the weighted mean of the probesets’ means and standard deviations, where the contributions to this first component were used as weights. The sign of this score was then corrected so that it was congruent to the sign of the probeset contributing the most to the first component.
Prior to merging the datasets, each of the expression matrices were standardized gene-wise using the GSE39582 dataset as a reference: first, we randomly selected a subset of samples from GSE39582 that matched as much as possible the frequency distribution in the target dataset regarding gender, age, stage, tumour location and MSI; then, expression values in the target dataset were centered and scaled according to the distribution observed in this subset sampled from GSE39582.
MSI status was imputed in each dataset separately using a published gene expression signature . For doing so, we summarized the signature as describe above; then a clustering analysis based on non-parametric density estimation was carried out on the resulting score as described in  and implemented in . Accuracy of this imputation was evaluated in dataset GSE39582, which included annotation of tumor microsatellite stability (96% and 81% accuracy for % MSS and MSI samples, respectively). Only MSS samples were kept for the final processed data leaving a total of 914 microarray samples available for analysis.
Five publicly available Affymetrix microarray datasets were downloaded from the NCBI GEO repository . These datasets included gene expression and clinical information from a total of 1.082 breast cancer patients. GSE1456  contains 159 samples from patients receiving surgery in the Karolinska Hospital of Stockholm (Sweden). GSE2034  includes data from 286 tumor samples of lymph-node-negative patients collected at the Erasmus Medical Center in Rotterdam (Netherlands). GSE2990  includes data from 189 invasive breast carcinomas treated at either the John Radcliffe Hospital in Oxford (UK) or the Uppsala University Hospital in Uppsala (Sweden). GSE3494  provides the expression profiling and survival information of 251 tumours archived at the Uppsala University Hospital in Uppsala (Sweden). Finally, GSE7390  contains the information of 198 untreated patients at the Bordet Institute in Brussels (Belgium).
The processing and normalization strategy described above for colon cancer samples was applied to breast cancer cohorts. Eklund metrics  and batches due to scan day were considered as adjusting covariates in a mixed effect model to remove expression changes due to possible technical artefacts.
The metabric for breast cancer data  was also downloaded but no extra data processing was undertaken. Each of the expression matrices from GEO were standardized gene-wise using the metabric dataset as a reference following the same procedure detailed for the CRC datasets.
ER classification (ER+ or ER-) was imputated using hierarchical clustering  from the expression of the ESR1 gene. Besides, HER2 classification (HER2+ or HER2) and PR classification (PR+ or PR-) were imputated using hierarchical clustering from the expression of the ERBB2 gene and the PGR gene, respectively. Only genes measured in all datasets from GEO and metabric were considered. Survival information (relapse event and months to relapse) was annotated as part of the KM plotter version 2010 . Only ER+ samples were kept for the final processed data leaving a total of 2.294 microarray samples available for analysis.
mcsurvdata package is loaded by
This package contains two ExpressionSet objects which can be accessed using the ExperimentHub interface:
eh <- ExperimentHub() dat <- query(eh, "mcsurvdata") nda.brca <- dat[["EH1497"]] nda.crc <- dat[["EH1498"]]
Survival information is available in attributes
tev (follow up time)
evn (event information 0 no event 1 event) for both brca and crc
data. Eklund metrics are also computed in attributes
rna.deg. Unique characteristics of the
tumors such as
msi information and
cms are annotated in the
colon cancer cohorts, whereas
HER2.status are annotated in the breast cancer cohorts.
## R version 4.1.0 (2021-05-18) ## Platform: x86_64-pc-linux-gnu (64-bit) ## Running under: Ubuntu 20.04.2 LTS ## ## Matrix products: default ## BLAS: /home/biocbuild/bbs-3.13-bioc/R/lib/libRblas.so ## LAPACK: /home/biocbuild/bbs-3.13-bioc/R/lib/libRlapack.so ## ## locale: ##  LC_CTYPE=en_US.UTF-8 LC_NUMERIC=C ##  LC_TIME=en_US.UTF-8 LC_COLLATE=C ##  LC_MONETARY=en_US.UTF-8 LC_MESSAGES=en_US.UTF-8 ##  LC_PAPER=en_US.UTF-8 LC_NAME=C ##  LC_ADDRESS=C LC_TELEPHONE=C ##  LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C ## ## attached base packages: ##  parallel stats graphics grDevices utils datasets methods ##  base ## ## other attached packages: ##  mcsurvdata_1.10.0 ExperimentHub_2.0.0 AnnotationHub_3.0.0 ##  BiocFileCache_2.0.0 dbplyr_2.1.1 BiocGenerics_0.38.0 ##  knitr_1.33 BiocStyle_2.20.0 ## ## loaded via a namespace (and not attached): ##  Rcpp_1.0.6 png_0.1-7 ##  Biostrings_2.60.0 assertthat_0.2.1 ##  digest_0.6.27 utf8_1.2.1 ##  mime_0.10 R6_2.5.0 ##  GenomeInfoDb_1.28.0 stats4_4.1.0 ##  RSQLite_2.2.7 evaluate_0.14 ##  httr_1.4.2 pillar_1.6.1 ##  zlibbioc_1.38.0 rlang_0.4.11 ##  curl_4.3.1 jquerylib_0.1.4 ##  blob_1.2.1 S4Vectors_0.30.0 ##  rmarkdown_2.8 stringr_1.4.0 ##  RCurl_1.98-1.3 bit_4.0.4 ##  shiny_1.6.0 compiler_4.1.0 ##  httpuv_1.6.1 xfun_0.23 ##  pkgconfig_2.0.3 htmltools_0.5.1.1 ##  tidyselect_1.1.1 KEGGREST_1.32.0 ##  GenomeInfoDbData_1.2.6 tibble_3.1.2 ##  interactiveDisplayBase_1.30.0 bookdown_0.22 ##  IRanges_2.26.0 fansi_0.4.2 ##  withr_2.4.2 crayon_1.4.1 ##  dplyr_1.0.6 later_1.2.0 ##  bitops_1.0-7 rappdirs_0.3.3 ##  jsonlite_1.7.2 xtable_1.8-4 ##  lifecycle_1.0.0 DBI_1.1.1 ##  magrittr_2.0.1 stringi_1.6.2 ##  cachem_1.0.5 XVector_0.32.0 ##  promises_188.8.131.52 bslib_0.2.5.1 ##  ellipsis_0.3.2 filelock_1.0.2 ##  generics_0.1.0 vctrs_0.3.8 ##  tools_4.1.0 bit64_4.0.5 ##  Biobase_2.52.0 glue_1.4.2 ##  purrr_0.3.4 BiocVersion_3.13.1 ##  fastmap_1.1.0 yaml_2.2.1 ##  AnnotationDbi_1.54.0 BiocManager_1.30.15 ##  memoise_2.0.0 sass_0.4.0
 Barrett, T. et al. NCBI GEO: Archive for functional genomics data sets - Update. Nucleic Acids Research 41, 991-995 (2013).
 Jorissen, R. N. et al. Metastasis-Associated Gene Expression Changes Predict Poor Outcomes in Patients with Dukes Stage B and C Colorectal Cancer. Clinical Cancer Research 15, 7642-7651 (2009).
 De Sousa E Melo, F. et al. Methylation of cancer-stem-cell-associated wnt target genes predicts poor prognosis in colorectal cancer patients. Cell Stem Cell 9, 476-485 (2011).
 Marisa, L. et al. Gene Expression Classification of Colon Cancer into Molecular Subtypes: Characterization, Validation, and Prognostic Value. PLoS Medicine 10 (2013).
 Laibe, S. et al. A seven-gene signature aggregates a subgroup of stage II colon cancers with stage III. Omics : a journal of integrative biology 16, 560-5 (2012).
 Gautier, L., Cope, L., Bolstad, B. M. & Irizarry, R. A. affy-analysis of Affymetrix GeneChip data at the probe level. Bioinformatics 20, 307-315 (2004).
 Bolstad, B. M. et al. Quality Assessment of Affymetrix GeneChip Data in Bioinformatics and Computational Biology Solutions Using R and Bioconductor (Springer, New York, 2005).
 Gentleman, R. C. et al. Bioconductor: open software development for computational biology and bioinformatics. Genome Biology 5, R80 (2004).
 Irizarry, R. A. et al. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4, 249-264 (2003).
 Jorissen, R. N. et al. DNA copy-number alterations underlie gene expression differences between microsatellite stable and unstable colorectal cancers. Clinical cancer research : an official journal of the American Association for Cancer Research 14, 8061-9 (2008).
 Azzalini, A. & Torelli, N. Clustering via nonparametric density estimation. Statistics and Computing 17, 71-80 (2007).
 Azzalini, A. & Menardi, G. Clustering via Nonparametric Density Estimation: The R Package pdfCluster. Journal of Statistical Software 57, 1-26 (2014). 1301.6559.
 Pawitan, Y. et al. Gene expression profiling spares early breast cancer patients from adjuvant therapy: derived and validated in two population-based cohorts. Breast Cancer Research 7, R953 (2005).
 Wang, Y. et al. Gene-expression profiles to predict distant metastasis of lymph-node-negative primary breast cancer. The Lancet 365, 671-679 (2005).
 Sotiriou, C. et al. Gene Expression Profiling in Breast Cancer: Understanding the Molecular Basis of Histologic Grade To Improve Prognosis. JNCI: Journal of the National Cancer Institute 98, 262-272 (2006).
 Miller, L. D. et al. From The Cover: An expression signature for p53 status in human breast cancer predicts mutation status, transcriptional effects, and patient survival. Proceedings of the National Academy of Sciences 102, 13550-13555 (2005).
 Desmedt, C. et al. Strong Time Dependence of the 76-Gene Prognostic Signature for Node-Negative Breast Cancer Patients in the TRANSBIG Multicenter Independent Validation Series. Clinical Cancer Research 13, 3207-3214 (2007).
 Curtis, C. et al. The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups. Nature 486, 346-352 (2012).
 Gyorffy, B. et al. An online survival analysis tool to rapidly assess the effect of 22,277 genes on breast cancer prognosis using microarray data of 1,809 patients. Breast Cancer Research and Treatment 123, 725-731 (2010).