(337g) Integrated Formulation and Process Design for 3D Printing of Pharmaceuticals

Pharmaceutical manufacturing has entered a phase of exciting transformation. The conventional batch production schemes that have been a fixture in the industry are giving way to more efficient, novel processes such as continuous manufacturing and 3D printing (3DP).¹ Partly driven by disruptions suffered in the pandemic, the industry has also realized the importance of developing distributed manufacturing networks for agile and flexible drug production. Recent studies have sought to combine these advances to build continuous mini-plants that can service an individual patient’s drug requirement on demand.^2,3 For economic viability, these mini-plants need to be designed with a high degree of automated manufacturing capability. A major thrust in this effort has been to build integrated product and process design (PPD) frameworks to simultaneously design the product properties and manufacturing conditions based on prespecified performance characteristics desired from the product. The central idea in PPD problems is to solve the inverse estimation problem, i.e., given the functionality desired, design the product to match specifications and design the process to manufacture it.^4,5

This study seeks to build an integrated PPD framework for a pharmaceutical 3D printing platform – Drop on demand (DoD) printing. DoD is a versatile inkjet 3DP that has been demonstrated to produce precise, personalized dosages for a variety of pharmaceutical formulations.^6,7 It also supports integration with upstream active ingredient synthesis steps and has been demonstrated to function as a drug product manufacturing unit of a continuous mini-plant.² Drug loading in the dosages produced by DoD need to be tightly controlled. Thus, the functionality desired in DoD is high printing consistency which is achieved when the drops generated have low variance in volume and have a low number of satellite drops. Printing consistency depends on both, formulation (product) properties like composition, viscosity, surface tension etc. and printing (process) conditions like drop ejection frequency, ejection pulse amplitude etc. For a novel formulation, the input (formulation and printing) conditions that give high printing consistency are conventionally determined experimentally. The PPD framework reported in this paper seeks to automate this process thereby engendering savings in time and material consumed.

To develop this framework however, a process model needs to be built to determine printing consistency for given input conditions. Theoretical studies modelling drop formation (occurrence of satellite drops, uniformity of drops generated etc.) in inkjets tend to focus on Newtonian fluid systems, however these cannot be directly applied to pharmaceutical suspensions where non-idealities such as polymeric carrier fluids and concentrated particle suspensions are frequently involved. Data driven approaches like machine learning can provide an alternate means to model the DoD printer.^8,9 There is a plethora of machine learning techniques available to model process operations, artificial neural networks are one of the most widely used among them.

Thus, as a prerequisite to building the PPD framework, this study seeks to develop a neural network based machine learning model for DoD. This model will predict a printing consistency metric for a given set of input conditions. To build this model, experimental data is generated for a variety of formulation and printing conditions. This model is then applied to build the PPD framework and answer the question: to achieve high printing consistency, what printing and formulation properties can be used. Data driven models give only a one-way relationship between the input variables and the output. This makes solving the inverse estimation problem challenging as specifying output (printing consistency) does not enable calculation of input conditions. To overcome this a two-stage approach is proposed: first, points in the input design space are sampled using a broad sampling scheme such as Latin hypercube. The model is evaluated at the sampled points, input conditions that give high printing consistency are then screened and used as input to the next stage. In the second stage, a series of heuristics are applied to select the most desirable solution from this set (figure 1).

In this paper we will demonstrate that this two-step approach can form an effective basis for a PPD framework for DoD that can automatically determine optimal formulation and printing conditions for manufacturing dosages of any given drug.

References:

1. Innovations in Pharmaceutical Manufacturing on the Horizon. National Academies Press; 2021. doi:10.17226/26009
2. Sundarkumar V, Nagy ZK, Reklaitis G v. Small-scale continuous drug product manufacturing using dropwise additive manufacturing and three phase settling for integration with upstream drug substance production. Journal of Pharmaceutical Sciences. Published online March 2022. doi:10.1016/j.xphs.2022.03.009
3. Adamo A, Beingessner RL, Behnam M, Chen J, Jamison TF, Jensen KF, Monbaliu JCM, Myerson AS, Revalor EM, Snead DR, Stelzer T, Weeranoppanant N, Wong SY, Zhang P. On-demand continuous-flow production of pharmaceuticals in a compact, reconfigurable system. Science (1979). 2016;352(6281). doi:10.1126/science.aaf1337
4. Eden MR, Jørgensen SB, Gani R, El-Halwagi MM. A novel framework for simultaneous separation process and product design. Chemical Engineering and Processing: Process Intensification. 2004;43(5). doi:10.1016/j.cep.2003.03.002
5. Ng KM, Gani R. Chemical product design: Advances in and proposed directions for research and teaching. Computers and Chemical Engineering. 2019;126. doi:10.1016/j.compchemeng.2019.04.008
6. Içten E, Giridhar A, Taylor LS, Nagy ZK, Reklaitis G v. Dropwise additive manufacturing of pharmaceutical products for melt-based dosage forms. Journal of Pharmaceutical Sciences. 2015;104(5). doi:10.1002/jps.24367
7. Radcliffe AJ, Hilden JL, Nagy ZK, Reklaitis G v. Dropwise Additive Manufacturing of Pharmaceutical Products Using Particle Suspensions. Journal of Pharmaceutical Sciences. 2019;108(2). doi:10.1016/j.xphs.2018.09.030
8. Brishty FP, Urner R, Grau G. Machine learning based data driven inkjet printed electronics: jetting prediction for novel inks. Flexible and Printed Electronics. 2022;7(1):015009. doi:10.1088/2058-8585/ac5a39
9. Wang J, Chiu GTC. Data-driven drop formation modeling in nanoliter drop-on-demand inkjet printing. In: ASME 2020 Dynamic Systems and Control Conference, DSCC 2020. Vol 2. ; 2020. doi:10.1115/DSCC2020-3292