Predicting Pharmaceutical Particle Size Distributions Using Kernel Mean Embedding

Validation of model-based design of experiments for continuous wet granulation and drying

This paper presents an application case of model-based design of experiments for the continuous twin-screw wet granulation and fluid-bed drying sequence. The proposed framework consists of three previously developed models. Here, we are testing the applicability of previously published unit operation models in this specific part of the production line to a new active pharmaceutical ingredient. Firstly, a T-shaped partial least squares regression model predicts d-values of granules after wet granulation with different process settings. Then, a high-resolution full granule size distribution is computed by a hybrid population balance and partial least squares regression model. Lastly, a mechanistic model of fluid-bed drying simulates drying time and energy efficiency, using the outputs of the first two models as a part of the inputs. In the application case, good operating conditions were calculated based on material and formulation properties as well as the developed process models. The framework was validated by comparing the simulation results with three experimental results. Overall, the proposed framework enables a process designer to find appropriate process settings with a less experimental workload. The framework combined with process knowledge reduced 73.2% of material consumption and 72.3% of time, especially in the early process development phase.

T-shaped partial least squares for high-dosed new active pharmaceutical ingredients in continuous twin-screw wet granulation: Granule size prediction with limited material information

This work presents a granule size prediction approach applicable to diverse formulations containing new active pharmaceutical ingredients (APIs) in continuous twin-screw wet granulation. The approach consists of a surrogate selection method to identify similar materials with new APIs and a T-shaped partial least squares (T-PLS) model for granule size prediction across varying formulations and process conditions. We devised a surrogate material selection method, employing a combination of linear pre-processing and nonlinear classification algorithms, which effectively identified suitable surrogates for new materials. Using only material properties obtained through four characterization methods, our approach demonstrated its predictive prowess. The selected surrogate methods were seamlessly integrated with our developed T-PLS model, which was meticulously validated for high-dose formulations involving three new APIs. When surrogating new APIs based on Gaussian process classification, we achieved the lowest prediction errors, signifying the method's robustness. The predicted d-values were within the range of uncertainty bounds for all cases, except for d90 of API C. Notably, the approach offers a direct and efficient solution for early-phase formulation and process development, considerably reducing the need for extensive experimental work. By relying on just four material characterization methods, it streamlines the research process while maintaining a high degree of accuracy.

Exploring the effect of raw material properties on continuous twin-screw wet granulation manufacturability

Twin-screw wet granulation (TSWG) stands out as a promising continuous alternative to conventional batch fluid bed- and high shear wet granulation techniques. Despite its potential, the impact of raw material properties on TSWG processability remains inadequately explored. Furthermore, the absence of supportive models for TSWG process development with new active pharmaceutical ingredients (APIs) adds to the challenge. This study tackles these gaps by introducing four partial least squares (PLS) models that approximate both the applicable liquid-to-solid (L/S) ratio range and resulting granule attributes (i.e., granule size and friability) based on initial material properties. The first two PLS models link the lowest and highest applicable L/S ratio for TSWG, respectively, with the formulation blend properties. The third and fourth PLS models predict the granule size and friability, respectively, from the starting API properties and applied L/S ratio for twin-screw wet granulation. By analysing the developed PLS models, water-related material properties (e.g., solubility, wettability, dissolution rate), as well as density and flow-related properties (e.g., flow function coefficient), were found to be impacting the TSWG processability. In addition, the applicability of the developed PLS models was evaluated by using them to propose suitable L/S ratio ranges (i.e., resulting in granules with the desired properties) for three new APIs and related formulations followed by an experimental validation thereof. Overall, this study helped to better understand the effect of raw material properties upon TSWG processability. Moreover, the developed PLS models can be used to propose suitable TSWG process settings for new APIs and hence reduce the experimental effort during process development.

Partial least squares regression to calculate population balance model parameters from material properties in continuous twin-screw wet granulation

In the pharmaceutical industry, twin-screw wet granulation has become a realistic option for the continuous manufacturing of solid drug products. Towards the efficient design, population balance models (PBMs) have been recognized as a tool to compute granule size distribution and understand physical phenomena. However, the missing link between material properties and the model parameters limits the swift applicability and generalization of new active pharmaceutical ingredients (APIs). This paper proposes partial least squares (PLS) regression models to assess the impact of material properties on PBM parameters. The parameters of the compartmental one-dimensional PBMs were derived for ten formulations with varying liquid-to-solid ratios and connected with material properties and liquid-to-solid ratios by PLS models. As a result, key material properties were identified in order to calculate it with the necessary accuracy. Size- and moisture-related properties were influential in the wetting zone whereas density-related properties were more dominant in the kneading zones.

Evaluation of the influence of material properties and process parameters on granule porosity in twin-screw wet granulation

In recent years, continuous twin-screw wet granulation (TSWG) is gaining increasing interest from the pharmaceutical industry. Despite the many publications on TSWG, only a limited number of studies focused on granule porosity, which was found to be an important granule property affecting the final tablet quality attributes, e.g. dissolution. In current study, the granule porosity along the length of the twin-screw granulator (TSG) barrel was evaluated. An experimental set-up was used allowing the collection of granules at the different TSG compartments. The effect of active pharmaceutical ingredient (API) properties on granule porosity was evaluated by using six formulations with a fixed composition but containing APIs with different physical-chemical properties. Furthermore, the importance of TSWG process parameters liquid-to-solid (L/S) ratio, mass feed rate and screw speed for the granule porosity was evaluated. Several water-related properties as well as particle size, density and flow properties of the API were found to have an important effect on granule porosity. While the L/S ratio was confirmed to be the dictating TSWG process parameter, granulator screw speed was also found to be an important process variable affecting granule porosity. This study obtained crucial information on the effect of material properties and process parameters on granule porosity (and granule formation) which can be used to accelerate TSWG process and formulation development.

Use of machine learning in prediction of granule particle size distribution and tablet tensile strength in commercial pharmaceutical manufacturing

In the manufacturing of pharmaceutical Oral Solid Dosage (OSD) forms, Particle Size Distribution (PSD) and Tensile Strength (TS) are common in-process tests that are controlled in order to achieve the quality targets of the end-product. The Quality by Design (QbD) concept elaborates process understanding and sufficient controls. However, for older pharmaceutical products upscaled to commercial phase with Quality by Testing (QbT) approach, the knowhow of the product-specific critical parameters is often limited. In this study, two predictive machine learning (ML) models were used for a commercial tablet product, for which historical data of raw materials, production, in-process controls and condition monitoring were available. With the aforementioned data, the aim was to predict the PSD and the TS that indicate the product quality. The feature importance was used to evaluate the parameter importance for the PSD and the TS. Partial dependence, in turn, was used to estimate the parameter impact on the predicted TS. The study illustrates the capability of the ML models to bring additional value for commercial products through the enhanced product-related knowhow.

Artificial Neural Networks to Predict the Apparent Degree of Supersaturation in Supersaturated Lipid-Based Formulations: A Pilot Study

In response to the increasing application of machine learning (ML) across many facets of pharmaceutical development, this pilot study investigated if ML, using artificial neural networks (ANNs), could predict the apparent degree of supersaturation (aDS) from two supersaturated LBFs (sLBFs). Accuracy was compared to partial least squares (PLS) regression models. Equilibrium solubility in Capmul MCM and Maisine CC was obtained for 21 poorly water-soluble drugs at ambient temperature and 60 °C to calculate the aDS ratio. These aDS ratios and drug descriptors were used to train the ML models. When compared, the ANNs outperformed PLS for both sLBF_Capmul^MC (r² 0.90 vs. 0.56) and sLBF_Maisine^LC (r² 0.83 vs. 0.62), displaying smaller root mean square errors (RMSEs) and residuals upon training and testing. Across all the models, the descriptors involving reactivity and electron density were most important for prediction. This pilot study showed that ML can be employed to predict the propensity for supersaturation in LBFs, but even larger datasets need to be evaluated to draw final conclusions.

Analysis of the Effects of Process Parameters on Start-Up Operation in Continuous Wet Granulation

Toward further implementation of continuous tablet manufacturing, one key issue is the time needed for start-up operation because it could lead to lower product yield and reduced economic performance. The behavior of the start-up operation is not well understood; moreover, the definition of the start-up time is still unclear. This work investigates the effects of process parameters on the start-up operation in continuous wet granulation, which is a critical unit operation in solid drug manufacturing. The profiles of torque and granule size distribution were monitored and measured for the first hour of operation, including the start-up phase. We analyzed the impact of process parameters based on design of experiments and performed an economic assessment to see the effects of the start-up operation. The torque profiles indicated that liquid-to-solid ratio and screw speed would affect the start-up operation, whereas different start-up behavior resulted in different granule size. Depending on the indicator used to define the start-up operation, the economic optimal point was significantly different. The results of this study stress that the start-up time differs according to the process parameters and used definition, e.g., indicators and criteria. This aspect should be considered for the further study and regulation of continuous manufacturing.

Identification of Granule Growth Regimes in High Shear Wet Granulation Processes Using a Physics-Constrained Neural Network

The digitization of manufacturing processes has led to an increase in the availability of process data, which has enabled the use of data-driven models to predict the outcomes of these manufacturing processes. Data-driven models are instantaneous in simulate and can provide real-time predictions but lack any governing physics within their framework. When process data deviates from original conditions, the predictions from these models may not agree with physical boundaries. In such cases, the use of first-principle-based models to predict process outcomes have proven to be effective but computationally inefficient and cannot be solved in real time. Thus, there remains a need to develop efficient data-driven models with a physical understanding about the process. In this work, we have demonstrate the addition of physics-based boundary conditions constraints to a neural network to improve its predictability for granule density and granule size distribution (GSD) for a high shear granulation process. The physics-constrained neural network (PCNN) was better at predicting granule growth regimes when compared to other neural networks with no physical constraints. When input data that violated physics-based boundaries was provided, the PCNN identified these points more accurately compared to other non-physics constrained neural networks, with an error of <1%. A sensitivity analysis of the PCNN to the input variables was also performed to understand individual effects on the final outputs.