Demonstration of the Feasibility of Predicting the Flow of Pharmaceutically Relevant Powders from Particle and Bulk Physical Properties

Abstract

Purpose: Understanding and predicting the flow of bulk pharmaceutical materials could be key in enabling pharmaceutical manufacturing by continuous direct compression (CDC). This study examines whether, by taking powder and bulk measurements, and using statistical modelling, it would be possible the flow of a range of materials likely to be used in CDC. Methods: More than 100 materials were selected for study, from four pharmaceutical companies. Particle properties were measured by static image analysis, powder surface area and surface energy techniques, and flow by shear cell measurements. The data was then analysed and a range of statistical modelling techniques were used, to build predictive models for flow. Results: Using the results from static image analysis a model could be built which allowed the prediction of likely flow in a shear cell, which can be related to performance in a CDC system. Only a small amount of powder was required for the image analysis. Surface area did not add to the precision of the model, and the available surface energy technique did not correlate with flow. Conclusions: A small sample of powder can be examined by Static image analysis, and this data can be used to give an early read on likely flow of a material in a CDC system or other pharmaceutical process, allowing early intervention (if necessary) to improve the characteristics of a material, early in development.

Applications of Artificial Intelligence and Machine Learning Algorithms to Crystallization

“…A self-organizing map and a backpropagation feed-forward neural network were previously trained on eight micromeritics properties by Kachrimanis, Karamyan, and Malamataris with high prediction power for the flow rate on lactose, starch, and dicalcium phosphate hydrate, although tested only for larger particle size ranges. Recent advances with pharmaceutical relevance were made by Barjat et al, who used shear cell, static image analysis, surface area, and surface energy data on more than 100 materials to train a radial-basis SVM to regress and classify flow function coefficient values, a metric for the cohesiveness of a powder material. Their results indicate that the measured properties correlate to the flow behavior reasonably ( R 2 between 0.69 and 0.82) and lead to good flowability classification models (area under ROC curve 0.79–0.84).…”

“…Similarly, regression models, such as neural networks with four inputs and only five hidden nodes, could reduce the average absolute error from 40% to 9% in predicting the permeability of granular materials by incorporating a fines ratio not typically included in standard analytical modeling. 115 The optimal formulation estimate that complies with a set of requirements of other selected properties and save excipients can be obtained using numerous ML techniques, as comprehensibly reviewed recently by Gao et al 114 and Yang et al 262 Powder flow behavior, for example, is a key factor to assess the potential for the manufacturability of oral solid dosage forms in drug development (e.g., tablet), 263 which is attractive due to its fewer unit operations involved in the production and its convenience for the patient. 264 Flowability is a problematic property to measure and predict, especially in the context of small particle sizes and uncontrolled shape distributions, 263 electrostatic forces, 265 and varying surface properties.…”

“…115 The optimal formulation estimate that complies with a set of requirements of other selected properties and save excipients can be obtained using numerous ML techniques, as comprehensibly reviewed recently by Gao et al 114 and Yang et al 262 Powder flow behavior, for example, is a key factor to assess the potential for the manufacturability of oral solid dosage forms in drug development (e.g., tablet), 263 which is attractive due to its fewer unit operations involved in the production and its convenience for the patient. 264 Flowability is a problematic property to measure and predict, especially in the context of small particle sizes and uncontrolled shape distributions, 263 electrostatic forces, 265 and varying surface properties. 266 Predicting powder flow behavior opens possibilities to fine-tune the crystallization process with reduced risks in the feasibility of downstream processing and easier scale-up.…”

Predicting pharmaceutical powder flow from microscopy images using deep learning

“…Yu et al 17 used partial least squares analysis to generate numerical descriptors corresponding to particle shape and size, successfully estimating FFc and stating that the most important variables for the prediction of powder flow were the diameter descriptors of the particles and the aspect ratio, and therefore emphasized the importance of considering multiple descriptors to characterize powder flow. Most recently, Barjat et al 18 used statistical modelling to infer trends from numerical values calculated as a result of analytical methods, demonstrating the feasibility of the prediction of powder flow of pharmaceutical powders from particle physical properties.…”

Artificial Neural Network and Support Vector Regression Modeling for Prediction of Mixing Time in Wet Granulation

“…The algorithm seeks to maximize the margin of linear hyperplane separating different categories. SVR models have also been applied to many pharmaceutical articles, such as the analysis of drug quantities [24,25], the prediction of drug permeability through the skin [26,27], and the forecast of powders' flowability during continuous direct compression [28].…”