The prognosis of glioblastoma, the most malignant type of glioma, is still poor, with only a minority of patients showing long‐term survival of more than three years after diagnosis. To elucidate the molecular aberrations in glioblastomas of long‐term survivors, we performed genome‐ and/or transcriptome‐wide molecular profiling of glioblastoma samples from 94 patients, including 28 long‐term survivors with >36 months overall survival (OS), 20 short‐term survivors with <12 months OS and 46 patients with intermediate OS. Integrative bioinformatic analyses were used to characterize molecular aberrations in the distinct survival groups considering established molecular markers such as isocitrate dehydrogenase 1 or 2 (IDH1/2) mutations, and O⁶‐methylguanine DNA methyltransferase (MGMT) promoter methylation. Patients with long‐term survival were younger and more often had IDH1/2‐mutant and MGMT‐methylated tumors. Gene expression profiling revealed over‐representation of a distinct (proneural‐like) expression signature in long‐term survivors that was linked to IDH1/2 mutation. However, IDH1/2‐wildtype glioblastomas from long‐term survivors did not show distinct gene expression profiles and included proneural, classical and mesenchymal glioblastoma subtypes. Genomic imbalances also differed between IDH1/2‐mutant and IDH1/2‐wildtype tumors, but not between survival groups of IDH1/2‐wildtype patients. Thus, our data support an important role for MGMT promoter methylation and IDH1/2 mutation in glioblastoma long‐term survival and corroborate the association of IDH1/2 mutation with distinct genomic and transcriptional profiles. Importantly, however, IDH1/2‐wildtype glioblastomas in our cohort of long‐term survivors lacked distinctive DNA copy number changes and gene expression signatures, indicating that other factors might have been responsible for long survival in this particular subgroup of patients.© 2014 UICC

Motivation: Comprehensive analysis of genome-wide molecular data challenges bioinformatics methodology in terms of intuitive visualization with single-sample resolution, biomarker selection, functional information mining and highly granular stratification of sample classes. oposSOM combines those functionalities making use of a comprehensive analysis and visualization strategy based on self-organizing maps (SOM) machine learning which we call ‘high-dimensional data portraying’. The method was successfully applied in a series of studies using mostly transcriptome data but also data of other OMICs realms. Availability and implementation: oposSOM is now publicly available as Bioconductor R package. Contact: wirth@izbi.uni-leipzig.de Supplementary information: Supplementary data are available at Bioinformatics online.