Digital Seed Train Twins and Statistical Methods

Model-based concepts and simulation techniques in combination with digital tools emerge as a key to explore the full potential of biopharmaceutical production processes, which contain several challenging development and process steps. One of these steps is the time- and cost-intensive cell proliferation process (also called seed train) to increase cell number from cell thawing up to production scale. Challenges like complex cell metabolism, batch-to-batch variation, variabilities in cell behavior, and influences of changes in cultivation conditions necessitate adequate digital solutions to provide information about the current and near future process state to derive correct process decisions.For this purpose digital seed train twins have proved to be efficient, which digitally display the time-dependent behavior of important process variables based on mathematical models, strategies, and adaption procedures.This chapter will outline the needs for digitalization of seed trains, the construction of a digital seed train twin, the role of parameter estimation, and different statistical methods within this context, which are applicable to several problems in the field of bioprocessing. The results of a case study are presented to illustrate a Bayesian approach for parameter estimation and prediction of an industrial cell culture seed train for seed train digitalization. This chapter outlines the needs for digitalization of cell proliferation processes (seed trains), the construction of a digital seed train twin as well as the role of parameter estimation and different statistical methods within this context, which are applicable to several problems in the field of bioprocessing. The results of a case study are presented to illustrate a Bayesian approach for parameter estimation and prediction of an industrial cell culture seed, as an example for seed train digitalization. It has been shown in which way prior knowledge and input uncertainty can be considered and be propagated to predictive uncertainty.