18 Feb 2011
Roland Eils
Integrative Bioinformatics and Systems Biology (iBioS)
University of Heidleberg, Germany
Numerical simulations of signal transduction models based on differential equations heavily depend on choice of reaction parameters and initial concentration of the state variables. For unraveling the qualitative and quantitative behavior of such models typically parameter scans are performed to sample the complex parameter space. Unfortunately, in the typical case of a very high-dimensional parameter space such a parameter scan can only cover a minor fraction of the entire parameter space. Here, I will describe an approach partially overcoming the problem of undersampling the parameter space by classifying the shape of the predicted input-output behavior revealed by model simulations over various parameter ranges into different shape classes. This allows us to automatically classify the input-output behavior of model predictions for 100,000s of model runs into different phenotypic classes. This approach offers novel functional insights into systems even in the case of partially unknown reaction parameters. I will demonstrate how this approach can be used to study the regulation of Natural Killer (NK) cell activation in the immune system, in particular how the balance between activating and inactivating receptors signals is achieved on a molecular level.