Ndahiro Nelson, Ma Edward, Bertalan Tom, Donohue Marc, Kevrekidis Yannis, Betenbaugh Michael
Department of Chemical and Biomolecular Engineering, Johns Hopkins University, Baltimore, MD 21218, USA.
Department of Applied Mathematics and Statistics, Johns Hopkins University, Baltimore, MD 21218, USA.
iScience. 2025 Jul 3;28(8):112944. doi: 10.1016/j.isci.2025.112944. eCollection 2025 Aug 15.
Rapid, cost-effective biomanufacturing of products like therapeutics, materials, and lab-grown foods depends on optimizing cell culture media, a complex and expensive task due to the combination of components and processing variables. This is especially important for therapeutic production using mammalian systems like Chinese Hamster Ovary (CHO) cells, where long development timelines contribute to high drug costs. Using Bayesian optimization (BO), adapted for bioprocess applications, our method supports multiple parallel experiments and incorporates thermodynamics-based constraints on media solubility to ensure feasible medium formulations. The approach is validated both in-silico and in experimental bioreactor settings, showing improved product titers compared to classical design of experiments (DOE) methods. This work bridges machine learning and physical modeling to create a more data-efficient process optimization strategy. The integration of this method into biomanufacturing pipelines together with robotics-assisted bioreactors paves the way for automated bioprocess optimization and more rapidly available and affordable biotherapeutics.
快速、经济高效地生物制造治疗药物、材料和实验室培养食品等产品,依赖于优化细胞培养基,由于成分和加工变量的组合,这是一项复杂且昂贵的任务。这对于使用中国仓鼠卵巢(CHO)细胞等哺乳动物系统进行治疗性生产尤为重要,因为漫长的研发时间导致药物成本高昂。我们的方法采用适用于生物过程应用的贝叶斯优化(BO),支持多个并行实验,并纳入基于热力学的培养基溶解度约束,以确保可行的培养基配方。该方法在计算机模拟和实验生物反应器环境中均得到验证,与经典的实验设计(DOE)方法相比,产品滴度有所提高。这项工作将机器学习与物理建模相结合,以创建一种数据效率更高的过程优化策略。将该方法与机器人辅助生物反应器集成到生物制造管道中,为自动化生物过程优化以及更快获得且价格更亲民的生物治疗药物铺平了道路。