Identification of Predictive Dynamic Models for Systems Biology

Wednesday, November 18, 2015 - 10:30am - 11:30am
Keller 3-180
Joerg Stelling (ETH Zürich)
Limited mechanistic knowledge, conflicting hypotheses, and still relatively scarce experimental data hamper the development of dynamic
models for cellular networks. In addition, systems identification is
often challenging because it concerns network topologies and parameters
simultaneously. This talk will address open problems and new methods for
systems identification at different levels of granularity, ideally
leading to detailed mechanistic systems models. To derive a
coarse-grained representation of network topologies, for example, to
identify unknown natural inputs to signaling pathways, we discuss a
probabilistic inference approach that relies on prototypic ‘network
motifs’ and its application to nutrient signaling in yeast. When a basic
model structure is inferred, but many mechanistic hypotheses are to be
evaluated, the method of ‘topological filtering’ enables one to
automatically generate a set of simple(r) models compatible with
observational data; this approach allowed us to identify a single,
highly plausible circuit topology in a stress signaling circuit by
experiment-theory iterations. Finally, we will emphasize how network
modularization and advanced numerical methods may achieve scalability of
such identification approaches, which is critical for many real-world