Conference Agenda

Pharmaceutical capacity planning is crucial to meet product demands from clinical to commercial stages. In recent years, the market boom of gene therapies and demand for vaccines in pandemic contexts has highlighted a need to shorten scale-up timelines and improve responsiveness of pharmaceutical supply chains to demand fluctuations and unforeseen events. To this end, the industry has seen an increasing uptake of single-use equipment (SU) to substitute more inflexible stainless-steel facilities (MU), allowing for rapid scale-up and scale-out of manufacturing capacity. In this space, investment planning is challenged by a need to make scale-up decisions before processes are fully intensified and process capabilities known for certain. Furthermore, process uncertainty in early stages of planning is combined with an uncertainty in future demands. In this context, an overestimation of attainable production targets and sub-optimal demand forecasting can result in shortages and larger costs.

In this work, we consider the integration of early-stage process uncertainty and demand uncertainty in the investment planning problem and account for the different timescales of uncertainty. We develop a planning tool integrating of process uncertainty using adaptive robust optimisation (ARO) and demand uncertainty using stochastic programming. Our framework consists of a quantification step where we quantify process uncertainty and cost-related inputs to the optimisation, followed by an optimisation step. Given a set of demand scenarios and process realisations along the time horizon based on ARO, the optimisation selects network structures and investment into facilities as first-stage decisions. Production levels at each manufacturing node, transportation flows and shipments are scenario-dependent second-stage decisions. In networks implementing MU equipment, the selection of parallel lines and scale is considered a scenario-independent decision, hence capturing the inflexibility of the equipment and longer timelines for recourse actions. In SU-based manufacturing, these variables become second-stage decisions and depend on demand realisations.

The adoption of the ARO approach leads to conversative decisions in the first-stage of the time horizon, with 10-fold larger costs and lower inventory accumulated. Results highlight that SU manufacturing leads to lower expected manufacturing costs and a better adaptation after risk-averse decision-making. Instead, MU results in less flexibility to cater for demands, thus leading to larger expected costs. This highlights that shortening the set-up times for capacity expansion leads to more responsive supply chains. Furthermore, the integration of process uncertainty helps establish more robust initial capacity plans that mitigate shortage risks on early stages of planning.

Continuous manufacturing has emerged as a crucial innovation in pharmaceutical tableting production, offering significant advantages in efficiency, scalability, and tablet quality. Understanding the complex dynamics of this process is essential to ensure the quality of the product across the production line. Data-driven modeling offers the opportunity to gain more insight into these types of processes. This study explores the application of the Sparse Identification of Nonlinear Dynamics (SINDy) method to model these dynamics. SINDy is a nonlinear identification technique that can identify the process dynamics in the form of first-order differential equations using only experimental data.

In silico data was generated using a flowsheet model of a Continuous Direct Compression line developed in gPROMS. This approach provided the flexibility to simulate a wide variety of experimental conditions, producing the data needed to train the SINDy model. The mass flow rate of the API feeder was used as the control input, while blend uniformity and content uniformity were defined as the state variables. To incorporate the effects of the mass flow rate, the SINDy with control (SINDYc) algorithm was used. A series of step changes and pulse inputs, along with their corresponding responses were generated to train the model.

An exhaustive exploration of different candidate functions was conducted and the main hyper-parameter of this model (λ) was fine-tuned to achieve the optimal level of sparsity in the model. Choosing the appropriate data scaling technique was a key step to obtain a good model performance.

The results show that the SINDy method, particularly with careful tuning of hyperparameters and data preprocessing, can effectively capture the key dynamics of a continuous direct compression tableting line. Future work will focus on validating the model with experimental data and investigating the effect of noisy signals.

The pharmaceutical industry is increasingly leveraging digital technologies, such as modelling and optimisation techniques, to enhance the efficiency of drug development processes. However, existing approaches face key limitations: 1) the absence of a comprehensive system of models to predict blend properties and final product attributes based on raw component properties, process conditions, and formulation; and 2) a lack of large-scale optimisation frameworks to achieve desired product quality attributes by combining physical and data-driven models. This study proposes a novel modelling and optimisation framework tailored to develop directly compressed tablets to achieve optimal drug quality while minimizing time and costs.

The hybrid framework integrates mixture and process models, both mechanistic and data-driven, to predict key characteristics like particle size, shape distribution, flowability, tablet porosity, and tensile strength. These models are incorporated into a digital optimisation system that fine-tunes tablet formulation and initial process conditions to meet critical quality attributes (e.g. porosity >15%, tensile strength >2 MPa). The framework's optimisation capabilities are further enhanced through a physics-informed Bayesian optimisation algorithm, which combines experimental data from an automated tablet manufacturing and testing system with physics-based compaction models to optimize process conditions while significantly reducing the number of required experiments.

Incorporating an advanced automated tablet manufacturing and testing system, this framework demonstrates a self-driven, robotics-based approach to conducting experiments. The system is equipped with an automated dosing unit, a bespoke powder transportation unit, and a compaction simulator, enabling precise powder dispensing, tablet production, and subsequent testing of tablet properties (e.g. weight, dimensions, tensile strength). Integrated near-infrared spectroscopy measures blend homogeneity, and a sessile drop system analyzes tablet performance through liquid uptake and swelling kinetics. All processes are digitally and physically integrated, allowing real-time adaptation of process parameters.

The proposed system was validated through several case studies, achieving accurate predictions of new active pharmaceutical ingredients (APIs) and successfully meeting desired quality attributes with up to 60% fewer experiments compared to traditional methods. The high-throughput automated system significantly reduces manual intervention, enhances precision, and mitigates the risk of human error. By integrating data-driven machine learning with physics-based models, the framework enables rapid and efficient process design, representing a transformative advancement in tablet manufacturing and pharmaceutical development.

In 2023, the ICH Q13 guideline for the development, implementation, and lifecycle management of continuous manufacturing (CM), was implemented in Europe (ICH, 2023). It promotes quality-by-design (QbD) and quality by control (QbC) strategies as well as the appropriate use of mathematical modelling. This urges a harmonizing understanding across academia and industry for adoption of interpretable models instead of black-box models, especially when applied in Good Manufacturing Practice (GMP) regulated areas (Altrabsheh,2023). These models can be obtained employing surrogate reduced order modelling which offers an entirely data-driven mean to represent highly reliable yet computationally intensive models in a lower-dimensional space (Ierapetritou et al., 2017; Pantelides and Pereira, 2024).

Advancements in data-driven system identification techniques, such as Dynamic Mode Decomposition with control (DMDc), are generating new opportunities for computationally efficient and explainable model development in comparison with complex physics-based models (Schmid, 2022). To this end, we propose a comprehensive model development using DMDc to represent the complex dynamics of CM processes in a lower-dimensional space, disambiguating between underlying dynamics and actuation effects. Simulation data was collected using a digital twin based on an integrated twin-screw granulation process – fluidized bed drying process at the Diamond Pilot Plant (DiPP).

Our model demonstrates low computational complexity while effectively capturing nonlinear dynamics with significant improvements observed in the performance metrics (e.g., r² > 0.93 for mean granule size prediction) when compared with state-space models obtained with N4SID algorithm of MATLAB System Identification Toolbox and Sparse Identification of Nonlinear Dynamical systems with control. Finally, we developed a closed-loop workflow that seamlessly connects data exchanges between Python (DMDc), GAMS (MPC optimisation) & gPROMS using the packages gO:Python and GAMSPy where we evaluate the controller performance with setpoint tracking and disturbance rejection studies. Results indicated high accuracy in real-time monitoring and control of granule size.

This study offers a novel, interpretable control strategy for CM. By integrating DMDc with MPC, we provide a robust framework that aligns with ICH Q13. The results demonstrate the potential for real-time release testing, reduced reliance on end-product testing, and improved process control, supporting the adoption of CM in the pharmaceutical industry.

References

Altrabsheh, E., Heitmann, M., Steinmüller, P., Pastori Vinco, B., 2023. The Road to Explainable AI in GXP-Regulated Areas. ISPE, Pharmaceutical Engineering 43(1).

ICH Q13, 2023. ICH guideline Q13 on continuous manufacturing of drug substances and drug products

Ierapetritou, M., Sebastian Escotet‐Espinoza, M., Singh, R., 2017. Process Simulation and Control for Continuous Pharmaceutical Manufacturing of Solid Drug Products, In: Tekin, F., Schönlau, A. (Eds.), Continuous Manufacturing of Pharmaceuticals. Wiley, pp. 33–105. https://doi.org/10.1002/9781119001348.ch2

Pantelides, C.C., Pereira, F.E., 2024. The future of digital applications in pharmaceutical operations. Curr Opin Chem Eng. https://doi.org/10.1016/j.coche.2024.101038

Schmid, P. J. (2022). Dynamic Mode Decomposition and Its Variants. Annu. Rev. Fluid Mech., 54(1), 225–254. https://doi.org/10.1146/annurev-fluid-030121-015835

Thrombolysis is a medical treatment aimed at dissolving blood clots that obstruct blood vessels, impeding the delivery of oxygen and nutrients. Conditions related to blood clots, such as stroke, heart attacks, and pulmonary embolisms, can be life-threatening. While several drugs are available to treat heart attacks and pulmonary embolisms, alteplase is the only FDA-approved option for thrombolysis in acute ischemic stroke (AIS) [1]. Researchers are currently exploring other thrombolytic drugs as possible alternatives to alteplase.

We have developed mechanistic models that can assess the efficacy and safety of various thrombolytics, including urokinase, pro-urokinase (proUK), alteplase, tenecteplase, and reteplase, for intravenous treatment of AIS [2-4]. These models combine pharmacokinetics and pharmacodynamics with a local fibrinolysis model using one-dimensional convection-diffusion-reaction equations. This approach allows us to predict outcomes such as lysis completion time and risk of intracranial haemorrhage (ICH).

Moreover, studies have shown a synergistic benefit when tissue plasminogen activator (tPA) is combined with pro-urokinase (proUK) in in vitro experiments [5]. Our model has been used to examine the combination of intravenous tPA and m-proUK as a promising treatment for ischemic stroke [6].

When comparing the effectiveness of different drugs in monotherapy, we found that urokinase achieves the fastest clot breakdown but carries the highest ICH risk due to severe depletion of fibrinogen in the bloodstream. Tenecteplase and alteplase have similar thrombolytic efficacy, but tenecteplase offers a lower ICH risk. Reteplase, despite the slowest fibrinolysis rate, maintains fibrinogen levels in systemic plasma during treatment.

For combination therapy, our simulations indicate that the complementary mechanisms of tPA and m-proUK can achieve clot dissolution times comparable to tPA alone while maintaining fibrinogen levels. Varying dose combinations showed that increasing the tPA bolus significantly reduces fibrinogen levels but only moderately improves clot breakdown time. Conversely, higher doses and longer infusion times of m-proUK had a minimal impact on fibrinogen levels but greatly improved clot lysis time.

Future research will focus on optimising treatment protocols by adjusting tPA bolus, m-proUK dosage and infusion rates, as well as exploring additional drug combinations. These adjustments could potentially maximise the therapeutic benefits of both combination and monotherapy for treating ischemic stroke. The full scope of work, from mechanistic modelling to optimisation, will be presented at the conference.

[1] FDA, Center for Drug Evaluation and Research Approval Package for Ivermectin, 1996.

[2] Gu et al., Pharmaceutics, 2019, 11(3), 111

[3] Gu et al., Pharmaceutical Research, 2022, 39(1), 41-56

[4] Yang et al., Pharmaceutics 2023, 15(3), 797

[5] Gurewich, J. Thromb. Thrombolysis, 2015, 40(4), 480-487

[6] Yang et al., Computers in Biology and Medicine, 2024, 171, 108141

Pharmaceutical manufacturing is witnessing a major transition from traditional batch to continuous mode of operation. This is because continuous manufacturing (CM) brings several benefits to the pharmaceutical industry which include a smaller CM equipment footprint that results in increased controllability and reduced capital cost. Additionally, CM can alleviate the scale-up challenge and reduce the development time. However, there exists a lack of convenient tools for facilitating CM design and development with which the drug substance and drug product unit operations can be readily integrated for the overall evaluation of process and product performance. To that end, the Python-based PharmaPy framework was proposed recently to advance the design, simulation, and analysis of these continuous pharmaceutical processes. However, the initial library of models only addressed upstream drug substance processing. In this work, new capabilities which include drug product unit operations have been added to the PharmaPy framework that are crucial for the manufacture of final solid oral-dosage products. As a consequence, PharmaPy now enables the end-to-end study and optimization of the effects of the material properties of the drug substance on solid oral dosage products. This is essential for improving product quality and reducing costs in product development and manufacturing. The new capabilities of the PharmaPy platform are demonstrated with process modeling and simulation studies using the sequential-modular approach. The added process design capability includes unit operations such as feeders, blenders, and tablet press that can be integrated with the drug substance unit operations such as reactors, crystallizers, filters, and dryers. The platform allows the development of different mechanistic, data-driven, or hybrid models to study and compare final output to support computational efficiency and model accuracy. Sensitivity analysis can be performed on the integrated end-to-end simulator to identify the critical input variables (material properties, process conditions, etc.) that influence the product quality. These subsets of input variables are also crucial for the development of control strategies. The analysis lowers the complexity of the model by ranking the significant input variables. Finally, feasibility studies are conducted on the extracted influential input variables to characterize the process design space to achieve desirable output. The accuracy and effectiveness of the feasibility analysis are increased by using a surrogate model technique. The proposed enhanced PharmaPy package can now support decision-making from the early research and development stages through manufacturing.