A predictive modular approach to constraint satisfaction under uncertainty - with application to glycosylation in continuous monoclonal antibody biosimilar production

TL;DR

The paper introduces a modular, learning-based predictive filter for constraint handling under uncertainty, combining a predictive filter, online adaptive constraint margins, and a constraint-violation cost monitor to minimize the cost of soft constraint violations. The method can be attached to any controller and operates with minimal modification to predicted inputs, enabling real-time constraint satisfaction without requiring a priori uncertainty models. The authors validate the approach with a detailed glycosylation constraint case in continuous monoclonal antibody biosimilar production, demonstrating substantial reductions in violation costs and the ability to adapt margin estimates as uncertainties change. This work provides a scalable, data-driven module for safe, economical operation in complex bioprocesses and other uncertain systems where learning-based controllers are used.

Abstract

The paper proposes a modular-based approach to constraint handling in process optimization and control. This is partly motivated by the recent interest in learning-based methods, e.g., within bioproduction, for which constraint handling under uncertainty is a challenge. The proposed constraint handler, called predictive filter, is combined with an adaptive constraint margin and a constraint violation cost monitor to minimize the cost of violating soft constraints due to model uncertainty and disturbances. The module can be combined with any controller and is based on minimally modifying the controller output, in a least squares sense, such that constraints are satisfied within the considered horizon. The proposed method is computationally efficient and suitable for real-time applications. The effectiveness of the method is illustrated through a realistic simulation case study of glycosylation constraint satisfaction in continuous monoclonal antibody biosimilar production using Chinese hamster ovary cells, for which the metabolic network model consists of 23 extracellular metabolites and 126 reactions.

Figures (10)

• Figure 1: The constraint violation of $q_{Glc}$ (top), proxy-based prediction of $P_{G_0}$ performance at steady state (middle) and corresponding predicted economic loss $\bar{L}_k$ (below, described later) with feedback controller (blue), feedback controller with predictive filter with $\theta = 0$ (red) and $\theta = 0.1$ (yellow) in the presence of uncertainties including disturbances. The average economic losses over the considered period are 1.43, 1.06 and 1.28, respectively. The purple lines represent $q_{Glc,ref} = 1.67$$\text{pmol/(cell$\cdot$day)}$ (top figure) and the corresponding optimal value $P_{G_0,opt} = 35\%$ (middle figure). The considered economic loss includes both the cost of constraint violation and the cost of operating away from $P_{G_0,opt} = 35\%$.
• Figure 2: The learning-based predictive filter framework
• Figure 3: Schematic of a perfusion process.
• Figure 4: G0 level glycoform percentage $P_{G_0}$ as a function of glucose consumption rate $q_{Glc}$.
• Figure 5: Schematic of medium with varying glucose concentration.