Model-based prediction methods depict an important tool in the development of knowledge-driven decision tools, such as digital bioprocess twins, showing a potential for accelerated biopharmaceutical process development, for example through a meaningful reduction of the experimental effort. However, this requires reliable predictions, as informative as possible, taking into account all known aspects and uncertainty. Challenges in this context are limited amount of high-quality experimental data, process non-linearity, the complexity of the cell metabolism, of the model and the ability to address batch-to-batch variabilities and further sources of uncertainty. To enable more effective and model-based decision tools, methods for parameter estimation, prediction and model updating were investigated in this thesis addressing two different main aspects: Integration of a priori available knowledge and propagation of uncertainty. They were made in the context of cell culture expansion processes because this part of biopharmaceutical manufacturing entails particular challenges, both practically and from a modeling and prediction perspective. However, the developed and investigated methods and workflows are not restricted to this process phase. First, novel computational methodologies and workflows, required to perform these investigations were developed and introduced. Next, it was investigated how the integration of prior knowledge within the parameter estimation process effects prediction performance using a moving horizon estimation (MHE) approach, considering also the role of data used for model updating. Moreover, a novel probability-based method for parameter estimation, prediction and model updating of cell culture process models, which enables integration of prior knowledge and calculation of propagated uncertainty using a Bayesian approach, was introduced and combined with statistical analysis, response surface modeling and multi-objective optimization.weiterlesen