Bond Graph Semantics - Biophysically and Thermodynamically Consistent Integration of Physiology

Abstract

Simulating complex biological and physiological systems and predicting their behaviour under different conditions is currently challenging. Decomposing systems into smaller and more manageable modules (hierarchical modelling) can address this challenge, assisting both model development and simulation. Nevertheless, many existing computational models in biology are not reusable and therefore difficult to assemble into larger models. Even when this is possible, the resulting model may be inconsistent with the laws of physics or the behaviour of the real physiological system. This may lead to ill-posed models that are physically impossible, impeding the development of realistic complex models in biology. This issue can be addressed by employing a domain-independent, energy-conserving modelling approach – bond graphs – which inherently applies the laws of thermodynamics and physics. Additionally, annotating models with standard biological terms (semantics) aids in automating model composition. This thesis proposes a general, semi-automated methodology for composing biosimulation models, combining the bond graph approach with semantic annotations. In this thesis, I introduce a robust approach to automatically compose biosimulation models in a modular and hierarchical manner. The major benefit is that modellers can spend more time understanding the behaviour of complex systems and less time wrangling with model composition. To demonstrate the functionality of my automated model composition framework, I applied it to two popular XMLbased formats for describing models of biology: the Cellular modelling Markup Language (CellML) and the Systems Biology Markup Language (SBML). Specifically, I present a framework that can automatically convert SBML and CellML models into equivalent bond graphs. The new bond graph models of CellML and SBML models are easily mergeable, resulting in physically plausible coupled models. I have demonstrated the applicability of my developed framework on CellML models by generating bond graph modules of the EGFR pathway, Ras activation, and the MAPK cascade and automatically integrating them into one whole pathway. I have also applied the same procedure to SBML models of pyruvate distribution and the pentose phosphate pathway. These applications demonstrate the capability of the tools I have developed which I hope will assist modellers of the future to compose models of complex biological systems.