To find the design space (DS) of a pharmaceutical process, quantification of the “assurance of quality” for the product under development is required. In this study, latent‐variable modeling is combined with multivariate Bayesian regression to identify a subset of input combinations (process operating conditions and raw materials properties) within which the DS of the product will lie at a probability equal to, or greater than, an assigned threshold. Partial least‐squares regression is used to obtain a linear transformation between the original multidimensional input space and a low‐dimensional latent space. The input domain is then discretized on its lower dimensional representation and a Bayesian posterior predictive approach is used to quantify the probability that the critical quality attributes of the product will meet their specifications for each discretization point. The methodology is tested on two case studies taken from the literature, one of which involving experimental data. The ability of the proposed approach to obtain a probabilistic identification of the DS, while simultaneously reducing the computational burden for the discretization of the input domain and providing a simple graphical representation of the DS, is shown. © 2018 American Institute of Chemical Engineers AIChE J, 64: 2438–2449, 2018

中文翻译：

---

药品开发中的概率设计空间确定：贝叶斯/潜变量方法

为了找到制药过程的设计空间（DS），需要对正在开发的产品的“质量保证”进行量化。在这项研究中，潜变量建模与多元贝叶斯回归相结合，以识别输入组合（过程操作条件和原材料属性）的子集，产品DS的概率等于或大于指定的阈值。偏最小二乘回归用于获得原始多维输入空间和低维潜在空间之间的线性变换。然后将输入域的低维表示离散化，并使用贝叶斯后验方法来量化产品的关键质量属性将满足每个离散点规格的概率。该方法论是在两个来自文献的案例研究中进行测试的，其中一个涉及实验数据。显示了所提出的方法获得DS的概率标识，同时减少输入域离散化的计算负担并提供DS的简单图形表示的能力。©2018美国化学工程师学会 示出了所提出的方法获得DS的概率标识，同时减少用于输入域的离散化的计算负担并提供DS的简单图形表示的能力。©2018美国化学工程师学会 显示了所提出的方法获得DS的概率标识，同时减少输入域离散化的计算负担并提供DS的简单图形表示的能力。©2018美国化学工程师学会AIChE J，64：2438–2449，2018