NCCR and other Projects

Database of gene expression and clinical data

One serious shortcoming of most microarray studies is the lack of sufficient sample size. This may be partially addressed by pooling all existing data and by adopting methods from statistical meta-analysis. For many of these datasets, the expression or clinical data are not conveniently organized. This requires manual collection and interpretation, followed by standardization of vocabulary and data representation. We are setting up a Genomic Medical Database, that is a curated collection of RNA / protein quantification and associated clinical data. We also map corresponding probes on different platforms to each other to facilitate the integrated study of multiple data sets. The focus is on breast cancer and the usability of the clinical and survival data, but the concept could be generalized.




Discovery of cancer subtypes, prognostic and predictive profiles

In collaborations with different biomedical teams, we apply bioinformatics and biostatistical methods to discover new informative markers and profiles, which can improve treatment decisions by oncologists. We described a new subtype of breast cancer, which might be responsive to anti-androgen drugs, in collaboration with Dr. Herve Bonnefoi at Geneva Hospital and Dr. Richard Iggo at the University of St Andrews (Scotland). In a continuation of the project, we search for profiles predictive of the sensitivity of tumors to the two major drugs for systemic chemotherapy against breast tumors, anthracyclines and taxanes. In collaboration with Dr. Christos Sotiriou and the Institut Jules Bordet in Bruxelles we have discovered a method to determinate a "gene expression" grade index (GGI), that is a powerful prognostic factor and that could supplant histologic grade in clinical practice. We are currently aiming at further improving the prognostic system with additional informative marker genes and studying gene-gene dependencies to better understand the molecular and mathematical basis of the risk of metastasis.




Improved programs for analyzing gene expression and association with phenotypes

There are a number of useful techniques from classical statistical analysis, that are underused in bioinformatics because they are not implemented to be effectively combined between them and with modern techniques of significance analysis in the multiple testing setting. We are developing a software package (GlimEx) in which we are integrating such methods. The core of the project is a C code library that can be used in a flexible way to solve heavy computational tasks. The aim is to give to analysts a tool for efficient integrated analysis of large data sets from multiple sources, freeing them from a number of tedious tasks, so that they can concentrate on specific issues such as result interpretation and identifying bias and confounding. Interfaces are developed to allow for use of the library from within R, in combination with other R and Bioconductor packages. GlimEx is planned to include different types of models for response (survival, logistic, etc.), built-in meta-analytical capabilities, adjustment methods to account for other (clinical) factors, multiple-testing correction methods and cross-validation algorithms for testing predictive models.




Other software developments

MAMOT is a program we developed to allow for the use of simple hidden Markov models by non specialists and that we is being used in a collaboration with Dr. P. Bucher for generating models of Transcription factor binding sites.and for teaching purposes.