Merck is optimizing its biomanufacturing productivity and increasing predictability by combining data analytics and operations research, according to the company.
These optimizations reportedly increased production batch yields by 50–60%, reduced variability by 20%, and enabled each of its production lines to manufacture one extra batch per week. It generated approximately $200 million of extra revenue without requiring additional raw materials, energy, equipment, or space. Notably, 40% less energy is required to product the same volume of product.
To achieve this, Merck teamed with researchers at Eindhoven University of Technology (TU/e). One important step was to optimize the bleed-feed process. That alone increased yield by 82% per bioreactor, notes Merck.
In bleed-feed technology, part of the cell culture is extracted during the exponential growth phase and fed a special medium. This culture then seeds new bioreactor runs, thus prolonging the exponential growth phase and minimizing the need to set up additional batches.
Core challenge
However, “the core challenge was to identify the optimal timing. If we bleed-feed too soon, we do not get high yield. If too late, we might fail, so striking the right balance was the main business and scientific challenge,” Tugce Martagan, PhD, associate professor, TU/e, tells GEN.
Martagan and Merck colleagues Marc Baaijens, associate director and bioprocess engineering & innovation lead, and Bram van Ravenstein, associate director and operations lead for bacteriological production, explained the process in a recent paper.
To perform the operation at the optimal time, the fermentation process is monitored at discrete intervals. The time since the last bioreactor setup or bleed-feed, the cell culture growth rate, and the number of bleed-feed operations performed are each recorded. At each step, the decision is made to continue or to bleed-feed. Bleed-feed implementation takes six to nine months.
To optimize the bioreactor, they established ranges for critical process parameters using industry-scale bioreactors and actual production orders (within specified regulatory ranges). This takes two to twelve months.
Gradually, researchers began relying upon current data, discrete event simulation modeling and optimization rather than historical data. This enabled development of a proactive planning and scheduling optimization model, which provided better insights into production capacities, lead times, and costs, says the Merck team.
This project transports seamlessly to other facilities, so “changing mindsets to accept data-driven decision making was the biggest pivot point,” adds Baaijens.
The researchers, therefore, recommend accessing a multidisciplinary network of academic and corporate expertise for deep domain knowledge in operations research and AI, as well as the specific product. “It’s all about people processes and tools,” Martagan says.
The initial project was conducted at Merck’s animal health manufacturing facility in Boxmeer, Netherlands. Plans are being assessed to apply the lessons learned to Merck’s human biomanufacturing facilities. For further optimization, the team plans to develop more advanced prediction models for real-time inferences on bioreactor productivity for continuous biomanufacturing.
