Bayesian dynamic model selection of non-isolated cell signaling pathways

Abstract

A cell signaling pathway is composed of a set of sequential chemical reactions that take place within the cell. The information is transmitted throughout that pathway through concentration changes in the involved chemical species. The dynamics of those pathways can be studied using ordinary differential equation-based models that describe reaction kinetics. Such models should be calibrated using experimental measurements of one or more chemical species and evaluated according to a given criterion, thus allowing us to select a model that explains well the experiments. To this end, Bayesian methods are interesting criteria, since they assign a posterior probability of a model given experimental measurements, which enable us to choose a model with highest posterior probability. However, model selection methods in general assume that the modeled signaling pathway is a reasonably isolated system, a hypothesis that often is not true and might lead to model estimation errors. Therefore, in this project we propose the development, implementation and validation of a model selection method that takes into account the lack of isolation that is intrinsic to cuts of cell signaling pathways. To this end, we will depart from Bayesian computation methods that are already available, and will also use the reduction of the model selection problem to the feature selection problem that was recently developed by our group. Finally, we will test the new method on artificial instances and also on experimental measurements that are yielded in cancer cells studies at our lab. We expect that the proposed method will become a relevant tool in the studying of cell signaling pathways in many biological contexts, specially in Cancer Biology studies. (AU)