Big data analytics and advanced mathematical modelling to enhance chromatography performance

每种新批准药物的研发成本不断增加，市场竞争日益激烈，给治疗蛋白制造商带来了越来越大的压力。公司正在努力实现更快、更具成本效益的流程开发和更具生产力的制造流程。据报道，下游工艺占治疗药物生产成本的50%-80%(Somasundaram等人，2018年)，经常被称为制造工艺的瓶颈。由于层析过程是标准蛋白质纯化系列中使用的重点技术，因此它们是过程开发和强化研究的一个重要焦点。层析系统的过程开发依赖于启发式方法所支配的耗时的实验方法。设计质量方法旨在实施更系统的开发方法，更加强调基于工程和科学原则的过程理解和控制。机械模型可以通过快速调查替代配置来增加流程理解，并加速流程开发。因此，对制备层析过程进行机械建模是生物制药工业非常感兴趣的问题，因此在这一领域的研究是至关重要的。为了降低与色谱生物分离相关的高昂操作成本，公司和学者也在研究过程分析技术及其在过程监测和控制中的应用。目前的工业控制系统通常使用简单的UV、pH和电导率信号来监控和响应产品质量的偏差。新的光谱技术与先进的多变量数据分析技术相结合，可能提供更好的在线产品质量估计，因此具有极大地增强层析控制的潜力。为了提高工业层析过程的性能，本研究的目的是开发和应用新型控制系统和机械建模到工业层析操作中。建议的工作如下：1.用于层析根本原因分析和过程性能改进的多变量数据分析。2.设计空间审问的机械模型开发与应用。3.可变光程紫外光谱作为增强层析控制的PAT。4.混合可变光程紫外光谱和机理模拟作为一种先进的层析控制方案。总而言之，新的过程分析技术和机械建模具有极大地提高工业色谱过程的稳健性和生产率的特点。通过使用相似的实验数据同时建立和验证经验多变量模型和机理模型，可以减少模型开发时间。此外，每种技术在应用于过程控制时的弱点都可以通过它们各自的优势来弥补，这可能会导致比单独应用这些技术更有效的层析过程。因此，对混合监控-建模控制系统的研究对生物制药行业具有重要意义。

The increasing cost of R&D per new approved drug and the growing competition in the market is placing increasing pressure on therapeutic protein manufacturers. Companies are striving for faster, more cost-effective process development, and more productive manufacturing processes. The downstream process is reported to account for 50-80% of the production costs of a therapeutic (Somasundaram et al., 2018), and is frequently referred to as the bottleneck of the manufacturing process. As chromatography processes are the focal technology used in a standard protein purification train, they are a great focus for process development and intensification research. Process development for chromatography systems relies on time-consuming experimental methods governed by heuristics. The Quality by Design approach aims to implement a more systematic development approach, with enhanced emphasis on process understanding and control founded on engineering and scientific principles. Mechanistic models can increase process understanding, and accelerate process development by enabling rapid investigation of alternative configurations. Therefore, the use of mechanistic modelling for preparative chromatography processes is of great interest to the biopharmaceutical industry, and so research in this area is critical. To reduce the high operating costs associated with chromatographic bioseparations, companies and academics are also investigating process analytical technology, and its applications in process monitoring and control. Current industrial control systems typically use simple UV, pH and conductivity signals to monitor and respond to deviations in product quality. Novel spectroscopic techniques, combined with advanced multivariate data analysis techniques, can potentially provide improved estimations of product quality in-line, and so possess the potential to greatly enhance chromatography control. Motivated by the desire to enhance the performance of industrial chromatography processes, the aim of this research is to develop and apply novel control systems and mechanistic modelling to industrial chromatography operations. The proposed sections of work are as follows: 1. Multivariate data analysis for chromatography root-cause analysis and enhancement of process performance. 2. Mechanistic model development and application for design space interrogation. 3. Variable pathlength UV spectroscopy as a PAT for enhanced chromatography control. 4. Hybrid variable pathlength UV spectroscopy and mechanistic modelling as an advanced chromatography control scheme. Together, novel process analytical technologies and mechanistic modelling possess qualities to greatly enhance the robustness and productivity of industrial chromatography processes. By building and validating the empirical multivariate models simultaneously with the mechanistic model using similar experimental data, model development times can be reduced. Furthermore, the weaknesses in each technology when applied for process control can be covered by their respective strengths, potentially resulting in a more productive chromatography process than if the techniques were applied separately. Thus, research into hybrid monitoring-modelling control systems is of great interest to the biopharmaceutical industry.