Mathematical modelling of biological systems facilitates a deeper understanding of organ function and disease mechanisms. However, biophysically based models have a complex structure, are computationally demanding and extremely difficult to validate, and comparison of competing models is very challenging. These characteristics seriously delay progress in the use of modelling in sciences such as systems biology and medicine, and as the application of models to both large basic science data sets and in clinical context progresses these issues will become increasingly important. The aim of the proposed project is to reduce these bottlenecks using metamodelling, i.e. generation of statistical approximations of model input-output mappings. The objectives are to 1) reduce computational demand of multi-scale spatiotemporal heart models by substitution of parts of the models with statistical approximations, 2) establish a robust platform for global high-dimensional sensitivity analysis (analysis of the impact of the various input parameters on the model outputs) based on metamodels, enabling more efficient model validation, and 3) develop metamodel-based methodology for model construction and validation through systematic comparison and assessment of the prediction spaces of different models and comparison of models to experimental data. We will build a flexible metamodelling framework based on multivariate regression, adapted to handling the high parameter- and state space- dimensionality characterising multi-scale models. The fellowship will have a large impact on integration of metamodelling into the modelling community due to the central role that the host group has in e.g. the FP7-funded Virtual Physiological Human project. The methodology will be generic and highly instrumental in the development and testing of complex models, and has the potential to make a major impact not only within the modelling community but also across both experimental and clinical sciences.