Dynamic optimization strategies for monoclonal antibody production

  • Dynamische Optimierungstrategien zur Produktion monoklonaler Antikörper

Kappatou, Chrysoula Dimitra; Mitsos, Alexander (Thesis advisor); Misener, Ruth (Thesis advisor)

Aachen (2020, 2021)
Book, Dissertation / PhD Thesis

In: Aachener Verfahrenstechnik series - AVT.SVT - Process systems engineering 11 (2020)
Page(s)/Article-Nr.: 1 Online-Ressource : Illustrationen, Diagramme

Dissertation, Rheinisch-Westfälische Technische Hochschule Aachen, 2020


Monoclonal antibodies (mAbs) constitute a high-value biopharmaceutical product with a wide range of applications, e.g., autoimmune diseases and cancer treatment. To meet the increasing market demand, address the entrance of biosimilar products and align with the quality by design (QbD) principles, considerable research effort has been devoted to the development of model-based approaches for mAb-producing processes. The complex nature and the dynamics of the biological systems used in the production of mAbs lead to nonlinear and conconvex mathematical models. Thus, the resulting dynamic optimization problems aiming at an increased process performance are typically multimodal. Although deriving suboptimal local solutions can have negative economical and safety impacts, deterministic global dynamic optimization is not yet tractable for these models. In this thesis, different dynamic optimization strategies for process intensification of antibody production to overcome the limitations resulting from convergence to suboptimal local solutions are investigated and theory on deterministic global dynamic optimization of a specific class of surrogate models is established. First, utilizing a predictive energy-based model for mAb production, it is shown how incorporating biological process knowledge into the optimization problem formulation expedites the derivation of superior local solutions. Furthermore, model reformulation and reduction techniques are investigated to improve the numerical properties of the model and lead to reduced computational effort and increased production outcome. In a similar direction, the advantages of decomposing the optimization problem into smaller more flexible optimization tasks are illustrated. In a next step, optimization incorporating product quality aspects is investigated, as increased antibody production should coincide with meeting certain quality specifications. To this end, different dynamic optimization problems are introduced to examine the effect of process intensification on glycosylation. Then, process performance is maximized with simultaneous control on product quality. The results successfully illustrate an example of how model-based dynamic optimization can be employed for implementation of the QbD approach in biopharmaceutics. Finally, to guarantee convergence of an optimization-based operating strategy to an ε-optimal solution, theory for deterministic global dynamic optimization is developed for a specific class of nonlinear data-driven dynamic models, namely Hammerstein-Wiener (HW) models. The presented method exploits the specific structure of HW models, and thereby extends existing theory on global optimization of systems with linear dynamics. The solution strategy is implemented in our open-source global optimization software MAiNGO to numerically solve examples from offline and online optimization. Additionally, an example motivated from antibody production is solved to global optimality using the described methodology, highlighting the potential and also the limitations of deterministic global dynamic optimization for bioprocess optimization.