Complex event types for agent-based simulation

This thesis presents a novel formal modelling language, complex event types (CET s), to describe behaviours in agent-based simulations. CET s are able to describe behaviours at any computationally represented level of abstraction. Behaviours can be specified both in terms of the state transition rules of the agent-based model that generate them and in terms of the state transition structures themselves. Based on CET s, novel computational statistical methods are introduced which allow statistical dependencies between behaviours at different levels to be established. Different dependencies formalise different probabilistic causal relations and Complex Systems constructs such as ‘emergence’ and ‘autopoiesis’. Explicit links are also made between the different types of CET inter-dependency and the theoretical assumptions they represent. With the novel computational statistical methods, three categories of model can be validated and discovered: (i) inter-level models, which define probabilistic dependencies between behaviours at different levels; (ii) multi-level models, which define the set of simulations for which an inter-level model holds; (iii) inferred predictive models, which define latent relationships between behaviours at different levels. The CET modelling language and computational statistical methods are then applied to a novel agent-based model of Colonic Cancer to demonstrate their applicability to Complex Systems sciences such as Systems Biology. This proof of principle model provides a framework for further development of a detailed integrative model of the system, which can progressively incorporate biological data from different levels and scales as these become available.