Open Access Highly Accessed Research

A consensus prognostic gene expression classifier for ER positive breast cancer

Andrew E Teschendorff1*, Ali Naderi1, Nuno L Barbosa-Morais12, Sarah E Pinder3, Ian O Ellis4, Sam Aparicio15, James D Brenton1 and Carlos Caldas1*

Author Affiliations

1 Cancer Genomics Program, Department of Oncology, University of Cambridge, Hutchison/MRC Research Center, Hills Road, Cambridge CB2 2XZ, UK

2 Institute of Molecular Medicine, Faculty of Medicine, University of Lisbon, 1649-028 Lisbon, Portugal

3 Cancer Genomics Program, Department of Pathology, University of Cambridge, Hutchison/MRC Research Center, Hills Road, Cambridge CB2 2XZ, UK

4 Histopathology, Nottingham City Hospital NHS Trust and University of Nottingham, Nottingham NG5 1PB, UK

5 Molecular Oncology and Breast Cancer Program, the BC Cancer Research Centre, West 10th Avenue, Vancouver BC, V5Z 1L3, Canada

For all author emails, please log on.

Genome Biology 2006, 7:R101  doi:10.1186/gb-2006-7-10-r101

Published: 31 October 2006

Additional files

Additional data file 1:

Weighted average D-index over the test sets in the training cohorts as a function of the number of top-ranked genes in the molecular classifier. The corresponding log-rank test p values are shown for the two training cohorts with test sets, NKI2 (blue) and EMC (green), separately. Results are shown for two independent choices of training/test set partitions (realizations) within the training cohorts.

Format: PDF Size: 32KB Download file

This file can be viewed with: Adobe Acrobat Reader

Open Data

Additional data file 2:

The weighted average D-index over the three training cohorts NKI2, EMC and NCH is shown as a function of the incremental number of top-ranked genes in the overall molecular classifier. Weights were chosen proportional to the number of samples in each cohort. The ranking of the genes was determined by the global average Cox-score over the ten training-test set partitions and three training cohorts.

Format: PDF Size: 5KB Download file

This file can be viewed with: Adobe Acrobat Reader

Open Data

Additional data file 3:

The 52-gene optimal classifier.

Format: PDF Size: 20KB Download file

This file can be viewed with: Adobe Acrobat Reader

Open Data

Additional data file 4:

The D-index (and associated 68% CI) of prognostic separation for the hybrid prognostic index (HPI ~ SPI + MPI*; red) in the three external cohorts (A1) JRH-1, (B1) UPP and (C1) JRH-2. In all cases, the risk-ordering of samples by the HPI is determined by the average ranking induced by SPI and MPI*. Also shown is the D-index of the standard prognostic index (BLACK) in each of the three external cohorts. The corresponding log-rank test p values (in log10-space) of the SPI and HPI classifiers are shown for the cohorts (A2) JRH-1, (B2) UPP and (C2) JRH-2. The 0.05 confidence threshold line is indicated (green).

Format: PDF Size: 14KB Download file

This file can be viewed with: Adobe Acrobat Reader

Open Data

Additional data file 5:

The mean D-index and associated standard errors of prognostic separation for the 10 hybrid prognostic (HPIp; red) indices in the three external cohorts (A1) JRH-1, (B1) UPP and (C1) JRH-2. In all cases, the risk-ordering of samples by the HPIp is determined by the average ranking induced by SPI and MPIp. Also shown is the D-index of the pathological/classical prognostic index (black) in each of the three external cohorts. The corresponding log-rank test p values (in log10-space) of the SPI and HPIp classifiers are shown for the cohorts (A2) JRH-1, (B2) UPP and (C2) JRH-2.

Format: PDF Size: 21KB Download file

This file can be viewed with: Adobe Acrobat Reader

Open Data

Additional data file 6:

The mean D-index and associated standard errors over the ten hybrid prognostic classifiers (HPIp) in the three external independent cohorts. The D-index of the HPIp is parameterized by the weight wM given to MPIp. Thus, for wM = 0 we have a pure histopathological classifier, while for wM = 1 the prognostic index HPIp = MPIp. It turns out that, because of the relatively small number of samples (36) with available node status and NPI information in JRH-2, there were weight values for which the D-index became too large for graphical representation.

Format: PDF Size: 32KB Download file

This file can be viewed with: Adobe Acrobat Reader

Open Data

Additional data file 7:

Gene ontology analysis results for the top 200 prognostic genes.

Format: GIF Size: 84KB Download file

Open Data

Additional data file 8:

Gene ontology analysis results for the top 200 prognostic genes.

Format: HTML Size: 62KB Download file

Open Data

Additional data file 9:

Gene ontology analysis results for the top 200 prognostic genes.

Format: TXT Size: 85KB Download file

Open Data

Additional data file 10:

Overlap of our 52-gene classifier with the prognostic signatures in [4,10-12].

Format: TXT Size: 1KB Download file

Open Data

Additional data file 11:

Clinical and gene expression data of the NCH cohort.

Format: GZ Size: 5.8MB Download file

Open Data