Qualitatively modelling genetic regulatory networks: Petri net techniques and tools

Abstract

The development of post-genomic technologies has led to a paradigm shift in the way we study genetic regulatory networks (GRNs) - the underlying systems which mediate cell function. To complement this, the focus is on devising scalable, unambiguous and automated formal techniques for holistically modelling and analysing these complex systems. Quantitative approaches offer one possible solution, but do not appear to be commensurate with currently available data. This motivates qualitative approaches such as Boolean networks (BNs) , which abstractly model the system without requiring such a high level of data completeness. Qualitative approaches enable fundamental dynamical properties to be studied, and are well-suited to initial investigations. However, strengthened formal techniques and tool support are required if they are to meet the demands of the biological community. This thesis aims to investigate, develop and evaluate the application of Petri nets (PNs) for qualitatively modelling and analysing GRNs. PNs are well-established in the field of computer science, and enjoy a number of attractive benefits, such a wide range of techniques and tools, which make them ideal for studying biological systems. We take an existing qualitative PN approach for modelling GRNs based on BNs, and extend it to more general models based on multi-valued networks (MVNs). Importantly, we develop tool support to automate model construction. We illustrate our approach with two detailed case studies on Boolean models for carbon stress in Escherichia coli and sporulation in Bacillus subtilis, and then consider a multi-valued model of the former. These case studies explore the analysis power of PN s by exploiting a range of techniques and tools. A number of behavioural differences are identified between the two E. coli models which lead us to question their formal relationship. We investigate this by proposing a framework for reasoning about the behaviour of MVNs at different levels of abstraction. We develop tool support for practical models, and show a number of important results which motivate the need for multi-valued modelling. Asynchronous BN s can be seen to be more biologically realistic than their synchronous counterparts. However, they have the drawback of capturing behaviour which is unrealisable in practice. We propose a novel approach for refining such behaviour using signal transition graphs, a PN formalism from asynchronous circuit design. We automate our approach, and demonstrate it using a BN of the lysis-lysogeny switch in phage A. Our results show that a more realistic asynchronous model can be derived which preserves the stochastic switch.