In a webinar roundtable discussion, scientists from leading pharmaceutical organizations joined heads to discuss how digital twins and mechanistic modeling are driving efficiencies in biomanufacturing processes. In this article we summarize the talking points and resulting discussion among Hooman Farsani, PhD, principal scientist at Pfizer; Associate Director of Data Analytics & Modeling Shruti Gopal Vij, PhD, from Takeda; Yuanyuan Cui, PhD, MSAT process engineer at Sanofi; and Sanofi Global Head of Process Data Management, ML and AI Platform, MSAT, Ramila Peiris, PhD.
To watch the webinar in full, view the on-demand version.
What are digital twins?
Integrating predictive modeling into bioprocess automation and control systems—or, creating a digital twin in the form of a digital simulation—will continue to fundamentally change the biopharma manufacturing industry. Optimized production and improved regulatory compliance are no longer optional, as biomanufacturers are constantly pressured to increase efficiencies and decrease time to market.
The panel discussion kicked off with defining digital twin and mechanistic modeling and expanding on how the two are related. Digital twins are replicas of physical systems that capture system variables and control the system based on natural laws. Mechanistic models are mathematical models that can be used to define natural phenomena and are a core part of digital twins. They are related in that digital twins incorporate a range of mechanistic, data-driven, and statistical models to explain and control the physical system. “It’s a replica of a physical system,” Farsani says. “A digital twin captures the majority of the phenomena that happens within the physical system, from biology, chemistry, and physics. Mechanistic models are mathematical models that explain natural phenomena, and they are based on natural laws.”
“The end goal here is to be able to implement these in real time to control a system or to modify a process automatically,” Vij says. “You want this system to be able to a mimic what you as a human being or a lab scientist would do in those situations when you face them in the real life.”
Improving efficiency with digital twin technology
The panel discussed use cases, or what are drivers of adoption, including optimizing existing processes, reducing development timelines, improving technology transfer and scale-up, and improving monitoring and control. One emphasis was the potential for optimization and improved efficiency in manufacturing facilities.
Digital twins can be most immediately applied to improving operational efficiency in process development (PD), maintenance, and cell culture management. Replacing the time and energy to upskill your staff with mechanistic modeling is key to driving efficiencies, Cui believes. Vij says that digital twins, as a form of artificial intelligence (AI), can and should be used for both research and production purposes. “I think the general trend that I’m noticing, at least in the last two years, has been that AI has been at the top of every CEO’s, every president’s list.”
When discussing the importance of data-driven models and using AI to improve digital twins, they asked themselves, how much is AI adding to the accuracy of digital twins? On one hand, mechanistic modeling data must exist before an AI/machine learning layer can be implemented. On the other, simulation methodology can improve the volume of data that can be used for training and validating the machine learning models. In other words, simulation methodologies and data augmentation technologies can make digital twins more accurate.
Advancing adoption of digital twins in manufacturing
The panel went on to consider factors that can slow adoption, including budget limitations, lack of staff expertise, and limited data access. Other challenges are around data management, IT infrastructure, getting corporate buy-in, and regulatory considerations. Data management is a key challenge, Peiris believes. “You have data coming from different systems, manufacturing execution systems, process historians. Having that data layer ready and contextualized so that you can use that data and feed that data into these models is fundamentally important.”
Finally, they talked about the importance of collaboration between PD, data, QA, and regulatory professionals. They encourage manufacturers implementing digital twins to engage with regulatory agencies like the US Food and Drug Administration (FDA) early, with an eye on overcoming regulatory obstacles, namely data variability and bias. Additionally, these interactions might shape future regulatory guidelines.
In summary, if manufacturers set realistic expectations and take a holistic approach, they can successfully implement digital twins. Collaboration, integrating AI, a strong IT infrastructure, and strict data management are key to harnessing the large potential of digital twins.
To watch the webinar in full, view the on-demand version here.