(337bm) Process Modeling and Optimization in Biopharmaceutical Manufacturing

With the advent of artificial intelligence, Industry 4.0, and digital twin technologies, the biopharmaceutical industry is witnessing remarkable progress in in-silico process development. These innovative tools leverage mathematical models and data integrations to create virtual representations of biomanufacturing processes, enabling enhanced process understanding and decision-making. In line with this trend, my research effort is dedicated to developing and applying modeling approaches for platform development, feasibility studies, process characterization, design space identification, and optimization. By harnessing the power of comprehensive simulations and predictive analytics, I aim to accelerate the digital transformation of development and manufacturing of biopharmaceuticals, offering increased efficiency, quality, and flexibility while reducing costs and footprints.

Firstly, to support process intensification and study the feasibility of new processes from an economic and sustainability standpoint, a fully integrated continuous biomanufacturing platform for monoclonal antibody (mAb) was developed in-silico. This developed platform accounted for the media and buffer preparation steps, with the incorporation of innovative technologies such as the intensified seed expansion, continuous perfusion, single-pass tangential flow filtration, multicolumn chromatography, and membrane chromatography. The base-case scenario, considering a bioreactor scale of 500 L, yielded a total cost of goods (COGs) of $102.2/g, 4865.6 kg of process water and 11.1 kg of consumables per kg mAbs. Subsequently, a scenario analysis was performed to investigate the impact of upstream mAb titer and bioreactor scale on the economics and footprints. With the developed models, bottlenecks of the continuous process were also identified, providing key operational information for implementation.

Focusing specifically on downstream, the second part of my research developed a data-driven model for continuous Protein A capture processes to substitute mechanistic models, and the derivative-free model was used to characterize the design space. To address the trade-off between model accuracy and computational complexity, a machine learning (ML) tool — surrogate-based feasibility analysis with adaptive sampling — was proposed. The incorporation of ML algorithms resulted in a significant computational time reduction of 36x. The study comprehensively investigated the effects of critical design and process variables on the design space using an active set strategy.

Lastly, the research combined machine learning and mechanistic models to construct a hybrid model for simulating hydrophobic interaction chromatography (HIC). To address the issue of limited understanding of the underlying adsorption mechanism, artificial neural networks (NN) were employed to bridge the gaps between the postulated mechanism and the real process. We proposed that the hybrid model could be constructed by combining a neural network with a simple but well-known isotherm. With a very simple NN structure, a 62% improvement in calibration accuracy and 31.4% in validation accuracy are observed, compared with mechanistic model. The extrapolation capability of the hybrid model is tested to ensure its generalizability. Process optimization is carried out as well to find the optimal operating conditions under product quality constraints using the developed hybrid model.

References

[1] Ding, C.; Yang, O.; Ierapetritou, M., Towards Digital Twin for Biopharmaceutical Processes: Concept and Progress. Cell Engineering Volume 11: Biopharmaceutical Manufacturing: Progress, Trends and Challenges, Ralf Pörtner, Springer [In Production].

[2] Ding, C.; Ardeshna, H.; Gillespie, C.; Ierapetritou, M., Process design of a fully integrated continuous biopharmaceutical process using economic and ecological impact assessment. Biotechnology and Bioengineering 2022, 119 (12), 3567-3583.

[3] Ding, C.; Ierapetritou, M., A novel framework of surrogate-based feasibility analysis for establishing design space of twin-column continuous chromatography. International Journal of Pharmaceutics 2021, 609, 121161.

[4] Ding, C.; Gerberich, C.; Ierapetritou, M., Hybrid model development for parameter estimation and process optimization of hydrophobic interaction chromatography. Journal of Chromatography A, 2023, 1703, 464113.