(540b) Discovery of Hybrid Kinetic Bioprocess Models Integrating Metabolic Network Constraints

Bioprocesses are widely used in the production of pharmaceuticals and chemicals. However, their full potential is rarely harnessed and their optimization remains challenging [1]. A reason for this is the poor understanding of underlying process kinetics which arise from complex biological interactions that are dependent on the environment, organism, and product [2]. Moreover, available data is limited, with often only few bioprocess runs and measurements reported per study. One possible way to overcome these limitations consists of formulating hybrid models that incorporate (bio-)physical knowledge into data-driven process models [3]. Hybrid-model frameworks have been shown to consistently outperform purely data-driven models for bioprocess optimization [3], yet how to optimally build them (i.e., which type of knowledge should be used and how much) remains unclear.

Metabolic networks are built from omics data of microorganisms based on the principles of mass and electron balance [4]. Their constraint-based optimization, i.e., flux-balance analysis, is a powerful tool for the prediction of unobserved metabolic fluxes of cells [5]. Databases of highly curated metabolic networks (e.g., BIGG [6]) provide a source of knowledge that could be incorporated into hybrid bioprocess models for the discovery of kinetic bioprocess models. Although initial studies exist [7,8], they focus on the parameterization of an existing model structure rather than model discovery, leaving a largely untapped potential.

Here we present a novel framework for integrating knowledge of metabolic networks into hybrid process models. The framework is designed to work with modern machine learning packages facilitating its integration with several hybrid modeling architectures. Optimality constraints of the metabolic network are incorporated via a flux balance analysis surrogate model that ensures feasibility of mass and electron balance and predicts (non-)observed nutrient uptake and byproduct formation rates.

To illustrate the performance of our framework, we apply it to a case study on protein production in Escherichia coli. The metabolic network constrained kinetic bioprocess model is trained and validated on experimental data. Moreover, we predict optimal bioprocess conditions and compare our framework to other state-of-the art bioprocess modeling approaches.

We believe that our framework could improve the reliability and accuracy of the generated bioprocess models and thus could increase the process efficiency and reduce the cost of products. It also opens up new avenues for the multi-scale modeling of bioprocesses to better support decision-making in industry.

[1] Schlögel, G., Lück, R., Kittler, S., Spadiut, O., Kopp, J., Zanghellini, J., & Gotsmy, M. (2024). Optimizing bioprocessing efficiency with OptFed: Dynamic nonlinear modeling improves product-to-biomass yield. Computational and Structural Biotechnology Journal, 23, 3651-3661.

[2] Helleckes, L. M., Wirnsperger, C., Polak, J., Guillén‐Gosálbez, G., Butté, A., & von Stosch, M. (2024). Novel calibration design improves knowledge transfer across products for the characterization of pharmaceutical bioprocesses. Biotechnology Journal, 19(7), 2400080.

[3] Schweidtmann, A. M., Zhang, D., & von Stosch, M. (2024). A review and perspective on hybrid modeling methodologies. Digital Chemical Engineering, 10, 100136.

[4] Orth, J. D., Fleming, R. M., & Palsson, B. Ø. (2010). Reconstruction and use of microbial metabolic networks: the core Escherichia coli metabolic model as an educational guide. EcoSal plus, 4(1), 10-1128.

[5] Orth, J. D., Thiele, I., & Palsson, B. Ø. (2010). What is flux balance analysis?. Nature biotechnology, 28(3), 245-248.

[6] King, Z. A., Lu, J., Dräger, A., Miller, P., Federowicz, S., Lerman, J. A., ... & Lewis, N. E. (2016). BiGG Models: A platform for integrating, standardizing and sharing genome-scale models. Nucleic acids research, 44(D1), D515-D522.

[7] Leppavuori, J. T., Domach, M. M., & Biegler, L. T. (2011). Parameter estimation in batch bioreactor simulation using metabolic models: Sequential solution with direct sensitivities. Industrial & engineering chemistry research, 50(21), 12080-12091.

[8] de Oliveira, R. D., Procópio, D. P., Basso, T. O., & Le Roux, G. A. (2022). Parameter estimation in dynamic metabolic models applying a surrogate approximation. In Computer Aided Chemical Engineering (Vol. 51, pp. 211-216). Elsevier.