Lourenço Ana Sofia, Schuster Tobias, Lopes João Almeida, Kirsch Annette
Research Institute for Medicines (imed.ULisboa), Faculty of Pharmacy, University of Lisbon, Av. Professor Gama Pinto, 1645-003, Lisboa, Portugal.
Global Analytical Technology Lab, Merck Healthcare KGaA, Frankfurter Straße 250, 64289, Darmstadt, Germany.
Eur J Pharm Sci. 2025 Jan 1;204:106976. doi: 10.1016/j.ejps.2024.106976. Epub 2024 Nov 28.
This study proposes a novel non-linear modelling approach to predict the dissolution profiles of extended-release tablets, by combining a full-factorial design, curve fitting to the dissolution profiles, and artificial neural networks (ANN), with linear regression methods, partial least squares (PLS) and multiple linear regression (MLR) as benchmarks. Hydroxypropylmethylcellulose (HPMC) and carboxymethylcellulose (CMC) grades, active pharmaceutical ingredient (API) lubrication, and compression force were chosen as DoE factors. The resulting batches were tested to obtain their corresponding dissolution profile, and a first-order dissolution equation was fitted to each profile. ANN, PLS and MLR were used to model and predict the tablet-specific constant k which then served to simulate dissolution profiles. This study demonstrates how non-linear methods, specifically ANN, outperform traditional linear models in predicting the complex interactions affecting drug release from extended-release formulations.
本研究提出了一种新颖的非线性建模方法，通过结合全因子设计、溶出曲线拟合以及人工神经网络（ANN），并以线性回归方法、偏最小二乘法（PLS）和多元线性回归（MLR）作为基准，来预测缓释片的溶出曲线。选择羟丙基甲基纤维素（HPMC）和羧甲基纤维素（CMC）的等级、活性药物成分（API）的润滑以及压片力作为实验设计（DoE）因素。对所得批次进行测试以获得其相应的溶出曲线，并对每个曲线拟合一阶溶出方程。使用人工神经网络（ANN）、偏最小二乘法（PLS）和多元线性回归（MLR）对片剂特定常数k进行建模和预测，然后用该常数模拟溶出曲线。本研究证明了非线性方法，特别是人工神经网络（ANN），在预测影响缓释制剂药物释放的复杂相互作用方面优于传统线性模型。
