Scientific publications and related contributions
We consider it an obligation to share our expertise, findings and also to provide access to this information, contributing to the growth in scientific knowledge. Therefore, we use as many formats as possible to achieve the best reach.
Peer-reviewed research articles
To reduce the experimental burden in mammalian bioprocess development we investigated the potential of hybrid models. To evaluate the transferability across scales, a hybrid model was developed with shake flask DoE data and applied to 15 L stirred tank bioreactor cultivations. Additionally, we investigated the applicability of intensified design of experiments (iDoE) to cover a design space with fewer experiments.
Single-pass Tangential Flow Filtration (SPTFF) is a fully continuous alternative to TFF. In this publication, we show that hybrid models trained on a single TFF experiment reliably predict the performance of various modes of SPTFF with varying number of membranes. Our findings allow for minimal product wastage during process development and enable Digital Twins to expand the gained knowledge to multiple filtration types.
Advanced online sensors for real-time monitoring of bioprocesses are promising tools with high potential to enhance process understanding and transparency. We investigated the capability of such a process analyzer, proton-transfer-reaction mass spectrometry (PTR-MS), for the exhaust gas in HEK293 cultivations. Herein, we developed a cell density soft sensor and identified a sensitive online indicator for glucose depletion, which can be used to set up new process control strategies to increase consistency.
Reducing practical experiments and rapidly finding the optimum process conditions for protein production are of high interest. We demonstrated how hybrid models based on a small set of experiments can be applied as Digital Twins. The derived simulations recommended further experiments to be performed to gain confidence about the best conditions in the design space.
The hybrid model for TFF was expanded to model multiple product components to include the influence of impurities on the process performance. We showed that the presented model can predict complex interactions and outperforms well-known mechanistic models.
To significantly reduce the required number of experiments for upstream process characterization, we highlight the combined concept of hybrid modeling and intensified design of experiments. Herein, we demonstrate a reduced experimental workload by more than 66%, saving time, raw materials and goods.
In this publication, we present a new hybrid model structure to predict the duration of crossflow ultrafiltration. We highlight the advantages of this approach compared to the film theory, show how it predicts batch and fed-batch filtrations and its use as a digital twin to evaluate the influences of various process parameters on ultrafiltration processes.
To outline the limitations and shortcomings of state of the art modeling techniques, we performed an extensive DoE study and compared the well-established response surface and black-box model methodologies with more advanced hybrid modeling. We demonstrate that hybrid models are superior to these techniques and possess advantageous features for implementing advanced process control tools.
In this publication, we present the complete workflow to develop an accurate biomass soft sensor, from process data collection to implementing the final model. By the additional use of an advanced 2D-fluorescence sensor, a deeper examination of the cells' metabolism was possible, facilitating deeper process understanding.
Herein, we deal with the established but disadvantageous state of the art techniques to calculate specific rates in upstream processes. Consecutively, we present a highly precise and robust method, which is not susceptible to analytical errors, enabling batch to batch comparability and closer process investigation.
Other scientific contributions
This article provides an overview of the current situation in the biopharmaceutical industry and gives an understanding of the advantages of incorporating data and process knowledge to achieve the highest possible benefits for the business.
Dealing with bioprocess characterization, Mark provides a comprehensive overview of the necessity and importance of this process task but also shortcomings incorporated in the current implementation, e.g., a high number of required experiments to gain sufficient process knowledge. To overcome these, implemented features of the Novasign toolbox, such as hybrid modeling and digital twin applications as well as intensified DoE, can be applied.
Mark presented how to efficiently combine process knowledge with process data into one beneficial hybrid model structure for bioprocess development. On the basis of our use cases, he pinpoints the huge saving of time, and how to use these models for soft-sensing and model predictive control for both up- and downstream.
In the course of this webinar, Mark presents the advantages of combining process data and process knowledge into one beneficial hybrid model structure for bioprocess development. On the basis of our use cases, he pinpoints the huge saving of time and thereto related expenses.