Biomanufacturing methods use live systems (e.g., bacteria and viruses) during the production processes. However, the use of live systems introduces several operational challenges, such as batch-to-batch variability in production yields and lead times. To address these operational challenges, a multidisciplinary team of researchers from MSD Animal Health and Eindhoven University of Technology collaborated over four years and developed a portfolio of optimization models and decision support tools [1]. These tools were aimed at improving biomanufacturing efficiency using a variety of operations research (O.R.) methodologies [2].
Tool 1: How to reduce setups. Bleed-feed is a novel technology that allows biomanufacturers to skip intermediary bioreactor setups. However, the specific time at which bleed-feed is performed is critical for success. If the bleed-feed is performed “too soon,” we would not achieve the maximum yield from the batch. When performed “too late,” we will miss the opportunity for bleed-feed and perform a costly setup. To optimize the bleed-feed decisions, the team developed a stochastic optimization model using Markov decision processes [2]. The real-world implementation of the bleed-feed tool resulted in an 85% higher yield per setup.
Tool 2: How to maximize the fermentation yield. In this project, the team conducted a limited number of industry-scale experiments to increase fermentation yield. However, the results obtained from each experiment were subject to an inherent randomness owing to the biological dynamics of fermentation processes. Therefore, we developed an optimal learning model based on a Bayesian design of experiments and identified the best process configuration at only eight industry-scale experiments. The real-world implementation of the new process configuration resulted in a 50% higher batch yield.
Tool 3: How to deal with planning and scheduling. Most biomanufacturing systems have “no-wait’” constraints between different production steps. The no-wait constraint requires a smooth flow of all products throughout the system. For this purpose, the team first developed a simulation model of MSD’s operations, which contained 48 products with 8,000 unique routings on 25 pieces of equipment. The model was validated with two years of historical data. Next, we used simulation-optimization to generate weekly production schedules (rhythm wheels). The real-world implementation allowed one extra batch per production line to be produced each week, leading to an additional 18 million euros of revenue per year.
Impact
Industry implementation at MSD Animal Health had a significant impact with up to 50% increase in batch yield and an additional revenue of 50 million euros per year. As a side benefit, our team obtained a 20% reduction in the standard deviation of the annual production yield (as a result of a data-driven, O.R.-based approach). The applications of O.R. methodologies is new to the biomanufacturing industry. As more companies like MSD embrace O.R., we believe this will significantly help the industry provide faster and more affordable access to new drugs.
References
- Martagan, T., Koca, Y., Adan, I., van Ravenstein, B., Baaijens, M. and Repping, O., 2021, “Operations research improves biomanufacturing efficiency at MSD Animal Health,” INFORMS Journal on Applied Analytics, Vol. 51, No. 2, pp. 150-163.
- Koca, Y., Martagan, T., Adan, I., Maillart, L. and van Ravenstein, B., 2021, “Increasing biomanufacturing yield with bleed-feed: Optimal policies and insights,” Working paper, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3659907.
(t.g.martagan@tue.nl)Tugce Martagan is an assistant professor in the School of Industrial Engineering at Eindhoven University of Technology, Eindhoven, The Netherlands.
(bram.van.ravenstein@merck.com)Bram van Ravenstein is associate director and operations lead, Bacteriological Production, at MSD Animal Health, Boxmeer, The Netherlands.
