Hybrid dynamic flux balance modeling approach for bioprocesses: an E. coli case study.

Abstract

In this study, we present a hybrid dynamic flux balance analysis (DFBA) model, combined with Partial Least Squares (PLS) regression, to simulate cell culture behavior in response to variations in media composition. DFBA models typically incorporate a stoichiometric matrix representing metabolic reactions, leveraging the pseudo-stationarity assumption to reduce the number of parameters, which in turn minimizes the risk of overfitting. Here, PLS regression is employed to define kinetic rate constraints within the DFBA model, capturing the dynamic and non-linear nature of reaction rates over different culture phases. An optimization approach identifies the minimal number of kinetic constraints required, ensuring model accuracy without excessive complexity. Our hybrid model is validated through simulation case studies using an E. coli system, demonstrating its effectiveness in adjusting to changes in initial media composition. The case studies reveal that the model's accuracy improves with a more detailed stoichiometric matrix, particularly when larger networks or more varied metabolic environments are present. Additionally, the hybrid DFBA-PLS approach provides a robust and scalable modeling framework adaptable to other bioprocesses, offering insights into medium composition effects and highlighting its potential for bioprocess optimization.