(223b) Enhancing Bioprocess Efficiency through Model-Based Optimization: A Proof-of-Value Study

Pharmaceutical manufacturing encompasses a broad range of processes. Recent transitions from batch to continuous manufacturing have paved the way for innovative model-based algorithms, such as advanced process control and soft sensors. However, many processes, including bioreactor operations, still rely on fed-batch mode and could benefit from these advancements. The process under investigation involves recombinant protein production of human superoxide dismutase (hSOD) using E. coli in a 20L bioreactor. During the growth phase, biomass concentration increases by adding substrate at an exponential rate and controlling reactor temperature. The production phase starts once the biomass concentration has doubled, at which point an inductor is added to initiate the product formation. The process parameters – temperature, substrate feed rate, and inductor feed rate – impact both biomass- and product concentration. Traditionally, these parameters are selected based on cell growth assumptions or data collected using classical designs of experiments (DoE).

This talk will introduce a model-based optimization strategy aimed at improving production efficiency and offering a systematic, resource-efficient alternative to conventional operating strategies. The approach is based on a dynamic process model that integrates kinetic equations with data-driven artificial neural network (ANN) components, parameterized using DoE data [1].

Offline component: In the first step, various objectives - such as maximizing product yield, enhancing economic profitability by considering substrate costs, or minimizing the time required to produce a specific amount of product (e.g., according to downstream units availability) - are formulated as objective functions. These functions, along with process and equipment constraints, define the optimization problem. Solving this problem yields optimal trajectories for temperature, substrate feed rate, and inductor feed rate, which are then applied as the optimal production recipe. In addition to providing greater flexibility in achieving different objectives on demand, the results show a promising improvement over the available DoE for all investigated objectives.

Online component: The optimal production recipe and its corresponding outputs represent an ideal scenario in the absence of plant-model mismatches or process disturbances. To apply this in real-world conditions, an online optimization strategy using a model predictive controller (MPC) is implemented. In each iteration, the MPC dynamically adjusts process settings to minimize the deviation between real and target process outputs, ensuring alignment with pre-computed reference trajectories despite disturbances, uncertainties, or model inaccuracies. Performance tests were conducted under various fault scenarios, including temperature controller shutdown or offset, and pump faults for substrate and inductor feeds. Fault start times and durations were systematically investigated across a broad range. Simulation tests were performed, and yields for the open-loop, MPC, and optimal scenarios were calculated and compared. These tests enabled a quantitative assessment of the optimization benefits, including improvements in yield and profit, along with a systematic evaluation of fault impacts. Specifically, the performance tests helped identify fault categories that the controller could compensate for, as well as those that required recalculation of the optimal recipe or batch termination.

In conclusion, the simulation study demonstrated notable potential for process optimization, including reduced production time and increased yield. These results lay the foundation for further investigation, with the next step focused on implementing the developed algorithms on real systems to evaluate their performance under practical conditions.

___________________________________________________________________________________________________________________________________________________________

[1] Bayer B, Striedner G, Duerkop M, Hybrid Modeling and Intensified DoE: An Approach to Accelerate Upstream Process Characterization. Biotechnology Journal. 2020, 15(9). doi: 10.1002/biot.202000121