# Accelerated model-based bioprocess development and process optimization

# â€‹

To characterize and optimize a bioprocess, for a newly developed drug or a biosimilar, is a necessary step in the biopharmaceutical industry before the product enters the clinical trials. Typically, this process characterization is achieved by using the statistical design of experiments (DoE) and process modeling. Hereby, critical process parameters (CPP) are defined at varying levels and experiments at different CPP combination settings are performed, setting up the so-called design space. The herein required number of experiments for a full-factorial characterization can easily exceed a feasible number, if too many CPPs and levels are chosen.

â€‹

number of required experiments = Levels^CPPs

â€‹

This makes design space characterization a time-consuming and laborious task. Consecutively, a hybrid model, based on this DoE data, can be developed using our toolbox and used for various process modeling purposes, e.g., time-resolved predictions of process variables of interest or digital twin applications. Even though this workflow leads to robust and reliable process models, still the shortcoming of a high experimental effort remains.

â€‹

To reduce this required number of experiments for process characterization, we propose the concept of intensified design of experiments (iDoE). Herein, intra-experimental shifts of the chosen CPPs lead to multiple characterized CPP combination settings within one bioprocess, i.e., a certain design space is completely characterized by performing fewer dynamic experiments, covering the complete design space.

â€‹

Figure 1. Reducing the number of experiments with intensified Design of Experiments (iDoE)

â€‹

Moreover, the combination of iDoE data and hybrid modeling results in the same advantages as using DoE data (with static process conditions) but the screening of particular design space is done more rapidly. Our showcase demonstrated that such an intensified hybrid model only required one-third of the data for model training (compared to a hybrid model developed using bioprocesses with static CPP setpoints), resulting in a reduced experimental effort of >66%, while maintaining high predictive performance. This highly emphasizes a combinatorial approach, of hybrid modeling and iDoE, which has the potential to accelerate bioprocess characterization.

â€‹

Figure 2. Predicting process variables of interest