We provide support to help achieve the goals of biomedical projects, no matter how big or small, through a wide range of activities. Our facility has acquired in-depth expertise in a broad range of studies and techniques, including cancer subtype discovery, biomarkers selection, class discrimination, cross-platform analysis, meta-analysis of multiple datasets. We participate in various research teams and international consortia, where we perform advanced data analysis tasks and methodological development or evaluation.

Cancer research

We enable, and participate in, projects with cutting-edge biomedical technologies. The typical aim is to advance translational research and individualized treatment for clinical patients, in particular, by finding biomarkers prognostic of patient survival or predicting treatment efficacy. Ongoing projects we support include Genomics and drug profiling of acute lymphoblastic leukaemia, characterization of the subtype structure of colorectal cancer, glioblastoma, rhabdomyosarcoma, and peripheral T-cell lymphomas. We test proposed biomarker signatures to clarify their utility; we assessed the currently strongest signatures that were proposed for clinical application in colorectal cancer patients to give a good picture of their relative merits and demerits.

  • An example: colorectal cancer:
    An investigation of gene expression data has revealed that colorectal cancer can be classified into at least four subtypes. After publication of our original study (Budinska et al. 2013, collaboration with CHUV, HUG, University Leuven and others), we have now reached, in an international consortium, an agreement on four “Consensus Molecular Subtypes” (CMS 1-4, Nature Medicine, in press). The four groups differ in molecular properties and in risk-level (probability of complete healing, patient survival duration in the case of metastases). There is a likelihood of improvement for patients (increased healing rates, prolonged average survival, reduced side-effects) by adapting the treatment to the subtype and developing new targeted treatments. These are the objectives of future research projects.

Bioinformatics and Biostatistics

We test data analysis methods to clarify their properties, so we compare methods for differential expression analyses with RNAseq data. We also write publications with a specific educational intent. For example, we have shown when and how confounding by batch effect or other causes can lead to strong bias in performance estimates, even when these are obtained by cross-validation on a learning set, compared to the true performance that can later be obtained on good independent data free of the same confounding effects.

  • An example: Rhabdomyosarcoma patients:
    Careful biostatistics modelling of patient survival data of Rhabdomyosarcoma (RMS) patients allowed us in collaboration with medical specialists (at The Institute of Cancer Research, Sutton, UK) to develop and propose for clinical use a new risk-classification system for RMS. For RMS there are no large datasets that could be used for this analysis, so we applied a cross-validation design to efficiently define a new classifier and at the same time test its performance without the risk of high over-fitting bias artefacts. The very strong difference between the groups in the plots demonstrates that the model carries much and very useful prognostic information. Finer analyses show that the new model is an improvement over older ones (Missiaglia et. al. 2012).

More information about our research







Example colorectal cancer projects (see “Cancer research")










The figure shows the Kaplan-Meier survival curves for the four Clinico-Molecular risk Groups we propose (CMG 1-4) for RMS (see “Rhabdomyosarcoma patients”)