Assessing the Impact of Patient Selection and Analytical Methods on mRNA Biomarker Identification in Ovarian Cancer Using TCGA Data.

IntroductionThe Cancer Genome Atlas (TCGA) data has been extensively used for differentially expressed genes (DEGs) discovery and validation in high-grade serous ovarian cancer (HGSOC), however platinum-sensitivity biomarkers have not yet reached clinical application. We aimed to propose a robust analytical framework for TCGA platinum sensitivity studies and systematically demonstrate how patient characteristics influence DEG analyses between platinum-sensitive and resistant groups, while also considering experimental platforms and analysis pipelines, as their impact is recognized but not comprehensively evaluated.MethodsThis retrospective TCGA cohort study used publicly available microarray and RNA-seq gene expression data. TCGA-derived datasets were identified through a literature review, including only studies enabling unambiguous patient identification. Three TCGA-based cohorts were assembled-two directly from published studies and one curated through own selection-differing in patient numbers (230, 201 and 142 patients, respectively) and stage distribution. Data were analyzed using DESeq2, edgeR, limma-voom, and Wilcoxon rank-sum test to compare the impact of patient selection and pipeline choice on DEGs outcome. Genes with an adjusted p-value < 0.05 and an |log2 fold change| > 2 were considered differentially expressed.ResultsThe three cohorts showed partial overlap of DEGs across platforms and pipelines. Both patient selection and analytical workflow influenced which genes were identified, highlighting findings variability even when using the same TCGA dataset. We establish a robust analytical framework for TCGA HGSOC biomarker studies, including the provision of patient barcodes and openly shared workflows that enable straightforward replication of our analyses from data download to DEG analysis.ConclusionsReproducibility of TCGA studies is limited by variability in patient selection, platinum sensitivity definitions, data sources, and analysis pipelines. We addressed these factors by providing a robust framework that can serve as a template for data analysis and transparent reporting when identifying and validating predictive signatures of platinum sensitivity in HGSOC patients.