Department of Pharmacy, University of Huddersfield, Queensgate, Huddersfield HD1 3DH, UK.
Department of Industrial Engineering, The University of Jordan, Amman, Jordan.
Int J Pharm. 2019 Sep 10;568:118542. doi: 10.1016/j.ijpharm.2019.118542. Epub 2019 Jul 19.
This study presents a modelling framework to predict the flowability of various commonly used pharmaceutical powders and their blends. The flowability models were trained and validated on 86 samples including single components and binary mixtures. Two modelling paradigms based on artificial intelligence (AI) namely, a radial basis function (RBF) and an integrated network were employed to model the flowability represented by the flow function coefficient (FFC) and the bulk density (RHOB). Both approaches were utilized to map the input parameters (i.e. particle size, shape descriptors and material type) to the flow properties. The input parameters of the blends were determined from the particle size, shape and material type properties of the single components. The results clearly indicated that the integrated network outperformed the single RBF network in terms of the predictive performance and the generalization capabilities. For the integrated network, the coefficient of determination of the testing data set (not used for training the model) for FFC was R=0.93, reflecting an acceptable predictive power of this model. Since the flowability of the blends can be predicted from single component size and shape descriptors, the integrated network can assist formulators in selecting excipients and their blend concentrations to improve flowability with minimal experimental effort and material resulting in the (i) minimization of the time required, (ii) exploration and examination of the design space, and (iii) minimization of material waste.
本研究提出了一个建模框架，用于预测各种常用药物粉末及其混合物的流动性。该流动性模型在 86 个样本上进行了训练和验证，包括单一组分和二元混合物。采用了两种基于人工智能（AI）的建模范例，即径向基函数（RBF）和集成网络，来对流动函数系数（FFC）和堆积密度（RHOB）表示的流动性进行建模。这两种方法都用于将输入参数（即粒径、形状描述符和材料类型）映射到流动特性上。混合物的输入参数是从单一组分的粒径、形状和材料类型特性确定的。结果清楚地表明，在预测性能和泛化能力方面，集成网络优于单一 RBF 网络。对于集成网络，测试数据集（未用于训练模型）的 FFC 决定系数 R=0.93，反映了该模型具有可接受的预测能力。由于混合物的流动性可以从单一组分的粒径和形状描述符来预测，因此集成网络可以帮助制剂师选择赋形剂及其混合物浓度，以在最小的实验工作量和材料浪费的情况下提高流动性，从而实现以下目标：（i）最小化所需时间，（ii）探索和检查设计空间，以及（iii）最小化材料浪费。
