Advances in Rule-based Modeling: Compartments, Energy, and Hybrid Simulation, with Application to Sepsis and Cell Signaling

Biological systems are commonly modeled as reaction networks, which describe the system at the resolution of biochemical species. Cellular systems, however, are governed by events at a finer scale: local interactions among macromolecular domains. The multi-domain structure of macromolecules, combined with the local nature of interactions, can lead to a combinatorial explosion that pushes reaction network methods to their limits. As an alternative, rule-based models (RBMs) describe the domain-based structure and local interactions found in biological systems. Molecular complexes are represented by graphs: functional domains as vertices, macromolecules as groupings of vertices, and molecular bonding as edges. Reaction rules, which describe classes of reactions, govern local modifications to molecular graphs, such as binding, post-translational modification, and degradation. RBMs can be transformed to equivalent reaction networks and simulated by differential or stochastic methods, or simulated directly with a network-free approach that avoids the problem of combinatorial complexity. Although RBMs and network-free methods resolve many problems in systems modeling, challenges remain. I address three challenges here: (i) managing model complexity due to cooperative interactions, (ii) representing biochemical systems in the compartmental setting of cells and organisms, and (iii) reducing the memory burden of large-scale network-free simulations. First, I present a general theory of energy-based modeling within the BioNetGen framework. Free energy is computed under a pattern-based formalism, and contextual variations within reaction classes are enumerated automatically. Next, I extend the BioNetGen language to permit description of compartmentalized biochemical systems, with treatment of volumes, surfaces and transport. Finally, a hybrid particle/population method is developed to reduce memory requirements of network-free simulations. All methods are implemented and available as part of BioNetGen. The remainder of this work presents an application to sepsis and inflammation. A multi-organ model of peritoneal infection and systemic inflammation is constructed and calibrated to experiment. Extra-corporeal blood purification, a potential treatment for sepsis, is explored in silico. Model simulations demonstrate that removal of blood cytokines and chemokines is a sufficient mechanism for improved survival in sepsis. However, differences between model predictions and the latest experimental data suggest directions for further exploration.