(568g) Crystallization Modeling of a Pharmaceutical Compound for Digital Twin Based in-Silico Optimization with Experimental Validation | AIChE

(568g) Crystallization Modeling of a Pharmaceutical Compound for Digital Twin Based in-Silico Optimization with Experimental Validation


Eren, A. - Presenter, Purdue University
Szilagyi, B., Purdue University
Quon, J., Takeda Pharmaceuticals
Papageorgiou, C. D., Takeda Pharmaceuticals International Co.
Nagy, Z., Purdue
The design of crystallization processes to achieve robust critical quality attributes (QCAs) is a practical challenge in the pharmaceutical industry. The current industrial standard still highly depends on Quality-by-Design (QbD), which requires intensive experimentation using traditional design of experiments (DoE) to find the operation conditions in the experimental space that satisfy the QCAs. With the current emerging Industry 4.0 practices the emerging paradigm has become the introduction of digital twins, which provide a predictive tool to be used in the optimization of the crystallization processes, or for in-silico DoE studies.1

In this work, we present the development and validation of a digital twin for the crystallization process of a pharmaceutical API (Compound A), from Takeda Pharmaceuticals International Co., which forms needle-like crystals. Several model training and validation experiments were performed with temperature cycling to promote the growth of the crystals for kinetic studies. For data acquisition and system monitoring during these experiments, process analytical technology tools (PAT) were used, including Mettler Toledo’s ParticleTrack G400 for in-line chord length distribution (CLD), for detecting the crystallization events such as nucleation and dissolution, as well as for the crystal count measurement, the Zeiss MCS621 ATR-UV/Vis spectrophotometer for concentration measurement, and the Mettler Toledo’s ParticleView v19 for in-situ microscope images. Additionally, Malvern’s Mastersizer 3000 Hydro MV was used as an off-line characterization tool to measure crystal size distribution (CSD) and the data from the PAT tools were directly used in the parameter estimation step.

Two novel features were introduced during the parameter estimation for digital twin development. First, to make the model parameter estimation more robust, the focused beam reflectance measurement (FBRM) data was directly used in the parameter estimation without further transformations. Secondly, a novel size dependent growth (SDG) rate expression was used in the model to capture the CSD dynamics considerably better than the standard SDG rate models. The developed digital twin includes secondary nucleation, SDG, and dissolution mechanisms, which were identified by the designed training experiments and the objective that required thermocycling for the desired CQAs. The aim was to grow the crystals as large as possible without exceeding a batch time of 40 h while having product crystals with bulk density values larger than 0.25 g/mL. For this purpose and from previous experience, the training and validation experiments were designed to have temperature cycles.2,3 The developed model was used for in silico DoE and process optimization and the simulation results were validated experimentally, demonstrating the benefits of model-based digital design for crystallization process development. In addition, the benefits of having a digital twin for in-silico optimization were demonstrated by comparing the findings from an experimental DoE work on Compound A to the ones from the optimized profiles, which were obtained by using the digital twin. The results demonstrated that the developed model was accurate and predictive enough to validate the general conclusions related to the required key features of the optimal temperature profiles identified by the experimental DoE. These results indicated that in order to obtain larger crystals with narrower CSD, multiple temperature cycles should be implemented at higher temperature ranges (50-65 °C). It was also shown that having a higher range for cycling with multiple cycles has a larger effect on product crystals size and bulk density than changing the seed type and loading.

The conclusions of the in-silico DoE and digital design were validated by experiments, demonstrating that the model was able to identify the key features of the optimal operating conditions for the crystallization of Compound A, when the aim was to produce large crystals by implementing temperature cycles. The orders of magnitude difference in the time required to reach this conclusion by in-silico and experimental DoE also demonstrated the benefits of having a digital twin of the process that can enable rapid crystallization process development by using a digital design framework.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported.


  1. Su, Q. et al. A perspective on Quality-by-Control (QbC) in pharmaceutical continuous manufacturing. Comput. Chem. Eng. 125, 216–231 (2019).
  2. Jiang, M. et al. Modification of crystal shape through deep temperature cycling. Ind. Eng. Chem. Res. 53, 5325–5336 (2014).
  3. Simone, E., Klapwijk, A. R., Wilson, C. C. & Nagy, Z. K. Investigation of the Evolution of Crystal Size and Shape during Temperature Cycling and in the Presence of a Polymeric Additive Using Combined Process Analytical Technologies. Cryst. Growth Des. 17, 1695–1706 (2017).
  4. Mydlarz, J. & Jones, A. G. On the estimation of size-dependent crystal growth rate functions in MSMPR crystallizers. Chem. Eng. J. 53, 125–135 (1993).