Analysis of Large-Scale Biological Networks with Constraint-Based Approaches over Static Models

Carito Guziolowski-Vargas

To this date many approaches exist that model a genetic regulatory network in order to elucidate the dynamics of the system. These methods focus, however, mainly on small-scale regulatory models. In this thesis we use a formal approach over qualitative large-scale regulatory networks, that models the equilibrium shift of the cell molecules between two steady states. We test the coherency between the network topology and gene expression data, by using a general interaction logical causal rule. The outputs of our approach are to measure the consistency of our data, diagnose inconsistent regions of the network with respect to the experimental data, and infer the qualitative variation of new network molecules. Our method reasons over the whole network of interactions using eefficient algorithms based either on decision diagrams, dependency graphs, or answer set programming. We proposed programs and bioinformatic tools that, based on these efficient implementations, automatize this reasoning. We validated this approach using the transcriptional networks of E. coli and S. cerevisiae, and the signaling network of the EWS-FLI1 human oncogene. Our main results were: (1) high prediction accuracy of the shifts of the network molecules, (2) effective manual and automatic corrections of the model and/or data, (3) automatic inference of the role of transcription factors, and (4) automatic reasoning over the causes that influence important phenotypes on a signaling network. All in all, we provided a methodology that can be applied to complete regulatory networks built at different molecular levels, by exploiting the constantly increasing high-throughput outputs.