Nonlinear Predictive Control of a Bioreactor by Surrogate Model Approximation of Flux Balance Analysis

Abstract

The application of economic model predictive control (EMPC) techniques in bioprocesses is scarce due to limitations in obtaining accurate dynamic models. Simplified unstructured models (e.g., Monod) can be easily developed, but their prediction capacity is poor. On the other hand, models based on dynamic flux balance analysis (dFBA) of the detailed metabolic network appear as a promising alternative. However, using dFBA inside the controller leads to a bilevel optimization problem, which could require excessive computational effort. In the present work, a control approach is proposed to tackle this limitation. By combining mass balances with a surrogate model for the metabolic network, our approach provides a significant reduction in online computation while still accurately describing the microorganism of interest. A case study of a fed-batch bioreactor of Saccharomyces cerevisiae for ethanol maximization was chosen to test the new approach. First, flux balance analysis simulations with Yeast 8.30 genome-scale model were performed, then a simple polynomial model was fitted to these data by partial least-squares regression (PLSR). The identification step showed that only 11 PLS components were necessary to allow the FBA to be replaced by the surrogate model with a good accuracy. The surrogate model was coupled to the EMPC, and the results were similar to those presented in the literature, where the bilevel optimization problem is explicitly solved. The EMPC was able to compensate for structural errors in the identification process, and it provided a higher ethanol titer in comparison to the open-loop operation. The results showed that applying surrogate models to dFBA is a viable strategy to the solution of a bilevel optimization problem.

Modeling With Uncertainty Quantification Identifies Essential Features of a Non-Canonical Algal Carbon-Concentrating Mechanism

“…Metabolic modeling in general has been highlighted as a particularly promising application of surrogate modeling, since metabolic modeling has biotechnological potential but is challenged by the complexity of metabolism and by the “trial and error” process which is often required to produce a working metabolic model (21). Surrogate modeling has found uses in dynamic flux balance analysis and process modeling for bioprocesses (51, 52). Our work expands on these investigations by demonstrating what is to our knowledge the first application of surrogate modeling to ODE-based compartmental modeling of biological systems.…”

Hybrid Physics-Informed Metabolic Cybergenetics: Process Rates Augmented with Machine-Learning Surrogates Informed by Flux Balance Analysis

“…However, our approach does not exclude the use of machine learning to refine uncertain parts of the model using process data, potentially in a synergistic manner. Furthermore, this work extends beyond previous polynomial-based functions for mapping sets of exchange metabolic fluxes predicted by FBA . Here, we explicitly consider and exploit the intracellular domain for cybergenetic control applications in dynamic metabolic engineering.…”

“…Furthermore, this work extends beyond previous polynomial-based functions for mapping sets of exchange metabolic fluxes predicted by FBA. 33 Here, we explicitly consider and exploit the intracellular domain for cybergenetic control applications in dynamic metabolic engineering.…”

Network Modeling and Control of Dynamic Disease Pathways, Review and Perspectives

“…Jabarivelisdeh et al [222] proposed a model predictive control algorithm on a reduced metabolic-genetic network model of E. coli to maximize ethanol production by adjusting the limited-oxygen supply to the E. coli. De Oliveira et al [223] proposed a model predictive controller to maximize ethanol production of Saccharomyces cerevisiae with the model based on a consensus yeast metabolic network by manipulating the glucose feed and the dissolved oxygen level. Hebing et al [224] used a robust multistage nonlinear model predictive control in a reduced metabolic network for Chinese Hamster Ovary (CHO) cells to maximize the cell concentration in the beginning of the process and the product concentration in the later phase by adjusting the pH value of the cell and the feed rate of the main substrate.…”