## Theoretical Projects

Systems biology models have a strong tendency to become large, since the underlying mechanisms usually
consist of a complex network of components and interactions, and because systems biology modelling usually
seeks to understand further information about these details. For this reason, models are typically
over-parametrised with respect to the available data. This implies that many of the predictions of a
model are the result of arbitrary choices among parameter manifolds, i.e. are infinitely uncertain and
of little interest. Nevertheless, some predictions of a model must be fulfilled if the model should be
able to explain the given data. For such uniquely identified model properties, we have coined the term * core prediction*.

For the identification of such core predictions, we are working with various approaches. One
approach uses modified optimization algorithms to approximate the space of acceptable parameter values.
Properties that are shared among this space are good candidates for being core predictions. Another
approach combines identifiability analysis with model reduction to achieve a reduced model with more
readily known identifiable properties. Further, the reduction is sought to be done in such a way that
the relations between the original and the reduced models are known. This allows for a translation, or *zooming*, between the models. The resulting zoomable model is referred to as a *core-box model*.

More specifically we work with methods concerning uncertainty estimation, model selection, decision support, multi-level modelling, and mixed-effect modelling.