Application of Machine-Learning Algorithms for Better Understanding of Tableting Properties of Lactose Co-Processed with Lipid Excipients

Abstract

Co-processing (CP) provides superior properties to excipients and has become a reliable option to facilitated formulation and manufacturing of variety of solid dosage forms. Development of directly compressible formulations with high doses of poorly flowing/compressible active pharmaceutical ingredients, such as paracetamol, remains a great challenge for the pharmaceutical industry due to the lack of understanding of the interplay between the formulation properties, process of compaction, and stages of tablets’ detachment and ejection. The aim of this study was to analyze the influence of the compression load, excipients’ co-processing and the addition of paracetamol on the obtained tablets’ tensile strength and the specific parameters of the tableting process, such as (net) compression work, elastic recovery, detachment, and ejection work, as well as the ejection force. Two types of neural networks were used to analyze the data: classification (Kohonen network) and regression networks (multilayer perceptron and radial basis function), to build prediction models and identify the variables that are predominantly affecting the tableting process and the obtained tablets’ tensile strength. It has been demonstrated that sophisticated data-mining methods are necessary to interpret complex phenomena regarding the effect of co-processing on tableting properties of directly compressible excipients.

Improving Tableting Performance of Lactose Monohydrate by Fluid-Bed Melt Granulation Co-Processing

“…This sample was also characterized with the lowest NWC, so the lowest amount of energy is stored in the compact and thus released upon the termination of compression pressure. Great lubricating properties of mixtures co-processed by melt granulation with GPS have been also recently reported by Djuris et al [22]. Despite the best lubricating properties, this sample did not fulfil the criteria for tensile strength, so it cannot be considered suitable for direct compression.…”

Applications of Artificial Intelligence and Machine Learning Algorithms to Crystallization

“…Tabletability encompasses more than flow behavior, however, extending to properties such as tensible strength and process-specific parameters like net compression, detachment, and ejection work. Djuris et al 268 used both coprocessed and physical mixtures of excipients and API to train a multilayer perceptron with a radial basis function and a Kohonen neural network. They regressed and classified these properties to build a prediction model for the tablet tensile strength.…”

Comparison of G-Means Algorithms and Kohonen Network in Solving Clustering Problems

“…Но обычно, данные алгоритмы используются как часть более сложной системы. К примеру алгоритм K-Means в работах [10][11] используется совместно с генетическим алгоритмом, а сеть Кохонена в [12][13][14] с персептроном.…”