Statistical methods for data integration into multi-scale models and uncertainty tracking in systems medicine of the liver

The research unit ‘Quantifying Liver Perfusion-Function Relationship’ (QuaLiPerF) aims at a holistic understanding of liver function in the context of liver resections. This subproject (P5) of the research unit QuaLiPerF applies Bayesian methods for data integration into multi-scale models and uncertainty tracking. This is done in cooperation with the modelling and experimental project partners, providing a stochastic way to track uncertainty in models on the cellular, lobular and later whole-body scale. Therefore, this project forms a framework around the other model projects of the research unit and should in the end yield to coupled models on multiple scales. To achieve this, statistical methods like Markov chain Monte Carlo (MCMC) sampling, Maximum Likelihood (ML) parameter estimation and machine learning will be applied. For example, by defining error functions for ODE models, making the optimization of a likelihood optimization problem feasible and allowing to transfer uncertainties in the data to uncertainties in the model output. Machine learning approaches could help to smartly select high impact parameters for individualized assessment of regeneration capacity and assessing the risk of liver failure. A further aim is to translate the models based on rodent data to models that can be used for patients in a systems medicine approach. This should ultimately lead to individualized models for human patients in the context of liver hepatectomy.