Generalization of Dynamic Metabolic Flux Models and their Application for Economic Optimization of Bioprocesses

生物制药市场正以前所未有的速度增长。例如，全球销售额从2009年的990亿美元增加到2017年的2280亿美元。中国人卵巢（CHO）细胞培养物生产的单克隆抗体（mAb）是该提案的重点应用之一，是前五名最畅销的产品。 为了满足这一需求，制药行业与FDA合作，正在积极寻求新的稳健设计、控制和优化策略。数学模型是这些策略的核心。当前申请的重点是开发CHO细胞的新型动态代谢通量模型（DMFM），并研究其在工艺优化中的用途。 拟议的建模方法和强大的优化算法，这些模型的基础上，将赋予一个独特的优势，加拿大制药工业在优化的培养基和细胞系的发展和强大的经济最佳操作的生物工艺。这些技术的应用将是竞争力和遵守监管限制的关键。 DMFM是基于受约束的约束优化的解决方案。据推测，DMFM需要较少数量的参数来描述数据，因此所得到的模型更不容易过度拟合。虽然DMFM已被研究用于细菌和藻类，但DMFM用于CHO的研究非常初步。由于DMFM基于基因组信息，并且鉴于哺乳动物细胞基因组的公认知识缺口，预期模型误差。因此，为了确保工业界采用所提出的技术，必须解决对模型误差的鲁棒性。 根据上述内容，该提案包括两个组成部分：1-开发哺乳动物细胞的DMFM模型，2-开发稳健的策略，其中模型逐渐学习，同时将过程推向最佳状态。 建模组件包括：i-哺乳动物细胞DMFM的合适的生物学目标函数的搜索，ii-热力学和遗传调控约束的列入和iii-在培养中的细胞的异质性的影响。这些模型将使用我们动物细胞实验室收集的数据进行验证。 优化组件将调查：i-用DMFM制定经济优化，ii-在更新模型的同时优化培养基的批次到批次方法，以及iii-基于DMFM的经济模型预测控制（EMPC）用于灌注和连续操作的调查。 该研究计划将培养和支持3名博士和1名硕士，他们将获得理论和实验技能的独特组合，这些技能在工业中具有非常高的需求，例如基于基因组的动态建模和鲁棒优化，细胞培养和过程分析技术。

The market for biopharmaceuticals is increasing at an unprecedented rate. For example, the world sales increased from US $99 billion in 2009 to US $228 billion in 2017. Monoclonal antibodies (mAb) produced in Chinese Hamster Ovary (CHO) cell cultures, which is one of the focal applications of this proposal, are the top five best-selling products. To respond to this demand, the pharma industry in collaboration with the FDA is actively seeking novel robust design, control and optimization strategies. Mathematical models are the heart of such strategies. The focus of the current application is to develop novel dynamic metabolic flux models (DMFM) for CHO cells and to study their use for process optimization. The proposed modelling approach and the robust optimization algorithms to be based on these models will confer a distinctive advantage to the Canadian pharma industry in the development of optimized culture media and cell lines and for robust economically optimal operation of bioprocesses. The application of these techniques will be the key for competitiveness and compliance with regulatory constraints. The DMFM is based on the solution of a constrained optimization subject to constraints. It has been postulated that DMFM requires a smaller number of parameters to describe data and thus the resulting models are much less prone to over-fitting. Although DMFM has been studied for bacteria and algae, the study for DMFM for CHO is very preliminary. Since DMFM are based on genomic information, and in view of the recognized knowledge-gaps of the mammalian cells' genome, model error is expected. Thus, to ensure adoption of the proposed techniques by the industry, robustness to model error must be addressed. Following the above, the proposal comprises two components: 1- development of DMFM models for mammalian cells and 2- development of robust strategies where the model is progressively learned while the process is driven towards the optimum. The modeling component includes: i-the search for suitable biological objective functions for DMFM of mammalian cells, ii-the inclusion of thermodynamic and genetic regulatory constraints and iii-the effect of heterogeneity of cells in culture . The models will be validated with data to be collected in our animal cell laboratory. The optimization component will investigate: i-the formulation of economic optimization with a DMFM, ii- Batch-to-Batch approach to optimize the media while updating the model, and iii- the investigation of the use of economic model predictive control (EMPC) based on a DMFM for perfusion and continuous operations. The research program will train and support 3 PhDs and 1 MSc who will acquire a unique combination of theoretical and experimental skills that are in very high demand in industry, such as genome based dynamic modeling and robust optimization, cell culturing and process analytical technologies.