Statistical modeling of drug response and pathway alterations

Coordinator:		Prof. Dr. Jörg Rahnenführer
Institution:		Fakultät Statistik, Technische Universität Dortmund
Homepage:		www.statistik.tu-dortmund.de

The overall research goal of the research network is to translate cancer genome discoveries into clinical practice. In the other subprojects a multitude of genomewide measurements will be generated, in particular genomewide patterns of genetic changes in tumor cells. In addition, silencing of selected signalling pathways will be performed on the cell lines in order to understand the role of the pathways in the presence of oncogenic mutations. The multivariate measurements of genetic changes in cancer cells will be linked to a measure of therapy success, for example drug resistance. In order to obtain optimal scores for determining crucial combinations of genetic changes and crucial steps in the signalling pathways appropriate statistical and algorithmic methods and tools are required. We will develop and provide such methodology in order to optimize therapeutic outcome for cancer patients dependent on genetic changes and altered molecular pathways. Two main methodical challenges will be addressed. We will identify the key players among the multitude of genetic changes observed in the cancer cells that have an impact on therapeutic outcome, and we will construct models of the relationships between the changes. This analysis requires statistical techniques of dimension reduction, i.e. the selection of the most relevant changes for predicting therapy success. The main goal is to predict response to targeted therapeutics. We will follow two complementary approaches. The first approach is divided into two steps. First, the number of changes is reduced to a small number of candidates. This choice will be based both on experiments of other subprojects and on statistical significance of the changes in a univariate test for correlation with therapeutic outcome. Then, models describing dependencies between the candidate genetic changes will be developed that optimally estimate therapy success. The second approach for dimension reduction combines both steps in a coherent way. It trades the interpretability of the first approach with a maximization of the expected prediction accuracy. The phenotypic outcome, here typically therapy success, is directly modeled as a function of all genetic changes, but the most important changes are implicitly simultaneously selected by favoring less complex models. Further, we will develop a methodology that evaluates the relationships between genetic changes in cancer cells and signalling pathways.


Administration:    Prof. Dr. Thomas Lengauer

Institute:   Max-Planck Institut für Informatik, Saarbrücken


We will implement a web-based user-friendly tool for prediction of therapy outcome based on genetic lesion profiles. The broad goal is to improve therapy choice for cancer and other diseases characterized by genetic alterations. We will create and maintain a database containing genetic, clinical, and phenotypic information on the tumors/ cell lines under study. Experimentally determined DNA copy number alterations and gene mutations are types of genetic data that will be available in this database. Another database will contain experimental data from silencing of selected signalling pathways. Then we will integrate the statistical techniques developed with the cooperation partners at the Dortmund University. These methods will be implemented and will be used for automated prediction of therapy outcome based on the genetic data provided in the data base. Finally, we will create a web server that provides assistance for therapy choice to the world-wide medical community, based on our prediction tool. In a first phase, the prediction will only use the genetic lesions data and later it will combine scores from the genetic profile and other clinical and histopathological data.