(183b) Superstructure Synthesis and Simulation-Optimization of Pharmaceutical Manufacturing Processes Using Pharmapy

During the development phase of bringing a new or existing pharmaceutical to market, various design and operational decisions must be made, which impact the cost and time of engineering development. Once a candidate molecule is chosen, a synthesis pathway must be identified where the consideration of cleaner processes and utilization of green chemistry pathways has gained traction over recent years. With the chemical synthesis pathway having been selected, process synthesis decisions must be made on what manufacturing route produces the product in the most economical way while adhering to strict quality guidelines. This challenge can present a significant hurdle due to the computational expense, modeling expertise, and accuracy required for the rigorous digital design of pharmaceutical manufacturing processes. Each of these decisions can be addressed by exploitation of binary and logic-based decision-tree structures to automate and evaluate a candidate compound, its synthesis, and ultimately its manufacturing pathways.

One framework that adequately addresses the synthesis problem statement is generalized disjunctive programming (GDP).¹ Here, process structural decisions are mapped to binary variables and a system of disjunctive logic statements define how selecting a particular option at each layer of the decision tree impacts other engineering decisions. In this case we consider that the candidate active pharmaceutical ingredient (API) and chemical synthesis pathway has been chosen beforehand, however the manufacturing route is yet to be determined. Considering the superstructure of a small molecule pharmaceutical manufacturing route, one may require a reaction step followed by a crystallization step for purification and subsequently a filtration step. If the reaction has multiple steps, we must choose if one reactor is sufficient, or more than one is necessary and whether intermediate separation steps are required. Also, there is a need to consider whether continuous or hybrid batch-continuous operating modes should be employed as improvements over the traditional batch processing mode. Thus, we should consider batch, semi-batch, and continuous operating modes for each of the units to understand the trade-off, while adopting the more controllable continuous manufacturing alternatives. Another option can be the fidelity of the model used for each unit.² For instance, in crystallizer design, a rigorous population balance model (PBM) solved using the finite volume method (FVM) enables tracking crystal size distribution (CSD) over time and capturing the effects of CSD on downstream operations. However, the computational time required for such a model is high when compared to the lower fidelity method of moments (MoM) model using only the moments of the distribution rather than the full CSD. The decision space can expand further, but it is apparent even with these few decisions for each unit that an organized logical decision framework, such as that employed in GDP, is necessary when performing simulation and optimization over a large number of alternative manufacturing routes.

Recently, PharmaPy has been developed as a tool for rapid in-silico comparison and simulation of pharmaceutical manufacturing pathways.³ There has been work reported employing both derivative-free and derivative-based optimization^3,4,5 methodologiesto execute simulation-optimization studies of processes modeled using PharmaPy. In this work, we utilize PharmaPy to simulate processes within an automated decision framework based on disjunctive logic to rapidly compare a large number of manufacturing pathways. The decision tree is pared down using heuristics that are translated into logic statements that significantly reduce the number of alternatives in the search space. These flowsheets are then simulated using a small set of pre-defined operating conditions. Then, those superstructure layouts that were either feasible or close-to feasible based on critical quality attributes (CQAs) and operating constraints were optimized using a derivative-free simulation-optimization method. It was found that fully batch systems tend to have larger design spaces, and thus had a much higher likelihood of adhering the critical quality attributes after the first round of simulation. The quality of the heuristics was also analyzed by performing the enumerated simulation analysis on both the full superstructure search space and the heuristically reduced search space. It was found that for the heuristics chosen in this work, most viable manufacturing routes existed in the reduced space, and thus the reduction in computation time by utilizing a heuristically reduced search space resulted in great benefit.

References

1. Türkay, M. and Grossmann, I. “Logic-based MINLP algorithms for the optimal synthesis of process networks”, Computers and Chemical Engineering, 20 (8):959-978, 1996.
2. Chen, Q. and Grossmann, I. “Modern Modeling Paradigms Using Generalized Disjunctive Programming”, Processes, 7, 839, 2019
3. Casas-Orozco, D., Laky, D., Wang, V., Abdi, M., Feng, X., Wood, E., Laird, C., Reklaitis, G.V., Nagy, Z.K. “PharmaPy: an object-oriented tool for the development of hybrid pharmaceutical flowsheets”, Computers and Chemical Engineering, 153, 2021.
4. Laky, D., Casas-Orozco, D., Laird, C.D., Reklaitis, G.V., Nagy, Z.K., “Simulation-optimization framework for grey-box optimization using PharmaPy”, Paper 182f, 2021 AIChE Annual Meeting, AIChE, 2021.
5. Casas-Orozco, D., Laky, D., Hur, I., Mackey, J., Mufti, A., Reklaitis, G.V., Nagy, Z.K., “Digital design of a Lomustine manufacturing process using PharmaPy”, Paper 317f, 2021 AIChE Annual Meeting, AIChE, 2021.