Optical and dielectric technologies allow the continuous monitoring of mammalian cells in a bioprocess without the need for manual sampling or labeling, as well as cell counts the new methods provide data on the state of the cells. For example, optical methods can determine the changing morphology associated with the loss of viability. Capacitance measurements can determine changes in cytoplasmic conductivity indicative of the first stages of nutrient deprivation. The combination of these methods allows fine control during biopharmaceutical production processes.  

Continuous bioprocessing not only needs automation to control conditions, but control strategies to optimize productivity. This holds true for mammalian and microbial expression systems, in which the metabolic load to the cells may increase over time. This contribution will show digital twin-based control strategies how to achieve optimum specific productivity of recombinant protein production.

At UCB, we employ a range of strategies for cell culture media optimization, and these have been evolved over time to minimize cost and resources and to meet important internal deadlines. The presentation will share our learnings from these experiences and recommend best practices for our industry colleagues working on similar projects.

Hybrid modelling tools were developed for in silico culture media design with minimal wet lab activity requirements. Hybrid flux balance analysis tools (Hybrid-FBA) were developed combining a genome-scale model (GEM) of CHO-K1 cells with empirical information deduced from principal component analysis (PCA) of historical data. A key advantage is the ability to learn from experience. While the GEM is a fixed part of the model, the PCA (and the hybrid ensemble per inherency) will improve with each new cultivation performed. After sufficient validation cycles, the developed tools can be used to design ab initio custom feeds for every new molecule.

Machine learning models used in mammalian cell cultures can leverage both historical and real-time data, with the final goal to ensure a consistent product output quality and optimised process operation. We will show how a state-of-the-art model can design optimal experimental setups in an active learning scenario, bringing an improvement both in the quality of the data produced by experiments, and in the yield and robustness of the process itself.

Quality by Design (QbD)-guided bioprocess development is cost effective only if knowledge is transferred from one project to the next (horizontal knowledge transfer) and one scale to the other (vertical knowledge transfer). We developed a novel hybrid machine-learning method to allow model processes across scales, projects, and products. This method retains and transfers knowledge, reducing experiments across scales and for the development of a process of a new product.

Sufficient process understanding in the QbD concept is a cumbersome task. Hybrid modeling and advanced design space screening methods reduce the required experimental effort to reduce overall development timelines. Conserving process understanding in these models enables model transfer from early-stage development until manufacturing. The presentation will cover model-assisted process characterization, process optimization, implementation of soft-sensors, and a model predictive control showcase. The audience will learn how to set up strategies for implementation of process modeling and digital twins and which pitfalls have to be overcome in order to accelerate the development from early-stage development until manufacturing. 

Following the launch of Eat Just's chicken nuggets, bioreactor-grown meat is now available at commercial outlets in Singapore. Other companies, using species such as Bos taurus or Sus scrofa, are developing technologies for manufacturing meat. Creating pork from a porcine cell line is more than simply using a high-intensity CHO-process to grow cells. This talk discusses challenges the nascent industry faces, learnings from the CHO-world, and where new thinking is needed if cell-based meats are to find space in supermarket aisles.

Bioprocesses for large-scale protein production are generally known to be labor-intensive, time-consuming and inefficient in the usage of assets. To ensure commercial viability of protein manufacturing for food application-- such as precision fermentation-derived caseins for human consumption – it is essential to rethink the entire bioprocess design, from GRAS strain engineering to lean downstream processing as well as a smart conceptual design that leads to a tremendous CAPEX reduction.

Biological additives (BAs), such as yeast extracts & peptones are important raw materials for mammalian & microbial growth media formulations.Selection & optimization of BAs, typically using Mixture DOE, can be enhanced using QSAR modelling. QSAR uses a multivariate summary of commercially available BA compositions. BA-QSAR is shown to improve selection &optimization of BAs & allows in-silico prediction of effects of manufacturing lot variability on cell culture performance.