A prognostic gene expression index in ovarian cancer—validation across different independent data sets

Abstract

Ovarian carcinoma has the highest mortality rate among gynaecological malignancies. In this project, we investigated the hypothesis that molecular markers are able to predict outcome of ovarian cancer independently of classical clinical predictors, and that these molecular markers can be validated using independent data sets. We applied a semi-supervised method for prediction of patient survival. Microarrays from a cohort of 80 ovarian carcinomas (TOC cohort) were used for the development of a predictive model, which was then evaluated in an entirely independent cohort of 118 carcinomas (Duke cohort). A 300-gene ovarian prognostic index (OPI) was generated and validated in a leave-one-out approach in the TOC cohort (Kaplan-Meier analysis, p = 0.0087). In a second validation step, the prognostic power of the OPI was confirmed in an independent data set (Duke cohort, p = 0.0063). In multivariate analysis, the OPI was independent of the post-operative residual tumour, the main clinico-pathological prognostic parameter with an adjusted hazard ratio of 6.4 (TOC cohort, CI 1.8-23.5, p = 0.0049) and 1.9 (Duke cohort, CI 1.2-3.0, p = 0.0068). We constructed a combined score of molecular data (OPI) and clinical parameters (residual tumour), which was able to define patient groups with highly significant differences in survival. The integrated analysis of gene expression data as well as residual tumour can be used for optimized assessment of the prognosis of platinum-taxol-treated ovarian cancer. As traditional treatment options are limited, this analysis may be able to optimize clinical management and to identify those patients who would be candidates for new therapeutic strategies.

TRRUST: a reference database of human transcriptional regulatory interactions

“…To test the benchmarking power of the TRRUST data, we used two inferred human TRNs: i) a published TRN inferred from a combined data set of ChIP-chip/seq from the hmChIP database33 and various related gene expression data using the ChIPXpress34 algorithm and; ii) an unpublished TRN inferred from a series of microarray samples from Gene Expression Omnibus (GEO)35 and GSE1476436 using the GENIE337 algorithm. The benchmarking power of a given set of TF-target interactions was assessed by their enrichment for each of successive bins of 1,000 inferred regulatory interactions, which were sorted by algorithm scores.…”

“…We combined those files to construct one example of a human TRN that comprises TF-target relationships ranked by score. We applied the GENIE3 algorithm37 to GSE1476436 microarray data to infer another example of a human TRN. More significant regulatory interactions have lower scores by the ChIPXpress algorithm and higher scores by the GENIE3 algorithm.…”

Prognostically relevant gene signatures of high-grade serous ovarian carcinoma

“…To verify the prognostic power of the CLOVAR survival signature, we used ssGSEA to analyze 64 TCGA expression profiles not included in the training dataset and 815 expression profiles with matching survival annotation from 6 published studies (9,10,23,(29)(30)(31). CLOVAR survival signature ssGSEA scores were calculated for each of the 7 validation datasets.…”

Gene-expression signatures in ovarian cancer: Promise and challenges for patient stratification

“…Many early generation ovarian cancer gene expression profiling studies focused primarily on the prognostic value of gene expression signatures. Results of these early prognostic gene expression studies [6][7][8][9][10][11][12][13][14] and those following [15][16][17][18][19][20][21][22][23][24][25] are summarized in Table 1 (Table 1). Most of these studies have identified a group of prognostically relevant genes in relatively small training sets but did, to their credit, validate the prognostic relevance of the respective gene signatures in independent cohorts.…”