Reusing simulation experiments for model composition and extension

As the number of available models as well as their complexity grows, model reuse becomes increasingly interesting. However, building new models based on reusing existing ones poses several challenges, and among others, checking validity of the newly built model and the model reuse is of paramount importance. The modeling process is typically accompanied by executing simulation experiments, e.g., for model validation. Therefore, this thesis aims to employ simulation experiments to support developing models via model reuse, with a focus on validating the resulting model. The essential idea presented by this thesis is to reuse simulation experiments conducted for model validation. Individual models are annotated with their simulation experiments that have been performed for validating the model. Upon reuse of those models for building new ones, the associated simulation experiments are also reused and executed with the new model, to inspect whether the key behavior exhibited by the original models is preserved or not in the new model. Hence, the changes of model behavior resulting from the model reuse are revealed, and insights into validity of the new model are provided. As a prerequisite, an explicit, declarative experiment specification containing four aspects is proposed, i.e., model configuration, simulation configuration, data processing method, and model behavioral property. The embedded domain specific language SESSL is exploited and extended for specifying experiments. The tool TAECS is developed to automatically adapt and execute experiments as well as analyzing the results. To allow an automated evaluation of experiment results, simulation-based model checking techniques are adopted and integrated. Two types of model reuse are considered, i.e., model composition and model extension. Reusing simulation experiments in supporting model development via composition and extension are investigated individually. The effectiveness of the developed approach for model composition is demonstrated by two case studies, a case study of Lotka-Volterra models and a case study based on replaying a realistic simulation study of Wnt/β-catenin pathway. Moreover, to illustrate the benefits of the developed approach for model extension, a case study of receptor ligand pathway is presented.