Digital Twins Drive Efficiency and Innovation: What You Can Learn From Pioneers in Life Sciences

Cognizant and CESPE embrace pharmaceutical sector to drive digital innovation

Some pharmaceutical companies already have digital twins in production at various sites, while others are aiming to meet a 2030 deadline. Still others are probing the possibilities. Cognizant and CESPE have gathered leading experts for a conversation about what is possible, and how to scale innovation. Their insights are beneficial across industries.

Copper mortars, century-old books and glass jars decorate the ornate wooden pharmacy cabinets in the large conference room of the Faculty of Pharmaceutical Sciences at Ghent University. Cognizant and CESPE, the Centre of Excellence in Sustainable Pharmaceutical Engineering & Manufacturing, could not have chosen a more suitable location for their exclusive roundtable discussion around digital twins. The contrast is almost palpable: old pharmaceutical techniques are the backdrop for presentations and discussions on cutting-edge technologies.

Together with CESPE, Cognizant aims to inspire the pharmaceutical sector and show what is already possible today with the right technology. This event is focused on life sciences, but the challenges and lessons highlighted today can resonate across industries. After all, digital twins as technology are in the top ten trends in manufacturing across industries. Who does not want more agility, flexibility and resilience in their organization?

Digital Twins?

A digital twin is a virtual replica of a component, system or process built from data. The twin helps to view, predict and understand real-world performance. This is a comprehensible, but also broad definition.

More specifically, a digital twin is a replica of the real world based on mechanistic information and data that is a representation of a process in the physical world. In the context of this roundtable discussion, think of a digital version of for example a chemical plant, a bioreactor, or a tablet production line.

Mechanistic vs. Data-Based Models

Mechanistic equations describe physical, biological and chemical processes as well as possible. They take physical laws into account but are complex and never really capture the complete picture due to inherent assumptions. All of reality does not fit in a usable equation (yet).

In contrast, data comes from all kinds of sensors and can be used to train models via machine learning. Such models produce very usable results but are time- and labor-intensive to produce. In addition, they are disconnected from the laws of physics.

"In practical terms, simplified hybrid models are the solution," says Prof Jan Verwaeren of Ghent University’s Department of Data Analysis and Mathematical Modelling. "These models use machine learning and data but still take relevant laws of physics into account."

Fermentation and Simulation

Dr. Matthieu Duvinage, Principal Data Scientist for Artificial Intelligence and digital twins at GSK, gives the example of a fermenter, in which a digital twin can initially help clarify what is happening inside and, in the next stage, also predict what is about to happen. The twin thus makes it possible to adjust parameters and choose the best way to proceed.

Professor Ashish Kumar, head of the Laboratory of Pharmaceutical Engineering in Ghent, further indicates the importance of digital twins in the optimization of complex manufacturing processes. "Often there are multiple manufacturing options available," the professor says. "Which method ensures the best delivery of a drug into the body? That requires many experiments, but when developing a new drug, the active pharmaceutical ingredient is very scarce and these experiments are expensive and time consuming."

A digital model of reality can be extraordinary valuable in more obvious cases. Professor Kumar continues, "Maybe you have made a good tablet, but now the marketing department suddenly wants a different shape and the change causes cracks. Such a question sounds simple but it is not, especially when you consider the time constraints." Professor Kumar uses his examples to emphasize an important point; he continues, "Digital models can help at all stages, from design and development to manufacturing and control."

Looking for the First Step

During the roundtable discussion, it turns out that not every organization knows how to start the process of digitalization with digital twins. Some companies are already undertaking large overall projects, but attendees are just as likely to talk about project-based ad hoc initiatives. An approach that enables rapid minimum viable products (MVPs) is the key, and for that Dr. Elisa Canzani, Data Science Lead at Cognizant, showcases a working solution.

"If you’re looking for really scalable technology, you need qualitative and automatic data flows.” GSK has engineered a solution for that with TwinOps, on which Dr. Canzani also wrote a paper. TwinOps is a platform consisting of building blocks for data ingestion, validation, transformation and storage. "Those building blocks can be reused after validation," she says.

TwinOps as the Foundation for Scalability

With the TwinOps platform in place, there is a sustainable and scalable link between data on one side and digital twins on the other. "Thanks to automated and repeatable workflows, it takes up to eighty per cent less effort to achieve compliance," says Canzani. The results are impressive: at GSK, the TwinOps platform enables engineers to arrive at an MVP in less than three weeks.

TwinOps is important not only for scalability, speed and compliance, but also to get the most out of digital twins. Canzani states, "A digital model without a data flow is not a true digital twin. Companies often have data, but it's in a colorful Excel." Only when data flows from sensors directly to models, the digital version of an asset truly becomes a digital twin.

Talking to Legacy

The data can flow both ways. The ultimate goal of most attendees is to connect models to their production environments. Data from bioreactors feeds the models, which offers insights based on simulations. Initially, these are then fed back to a human operator, but eventually such a smart system can adjust parameters itself and optimize production.

At this point, legacy tends to collide with digitization. Complex factories are controlled by software that is not always open. Several pharmaceutical companies struggle with the closed approach of their suppliers, with the control software being a bottle neck for digitization and automation. Getting data out of these control systems also proves to be quite the challenge for some. CESPE hopes to play a role here by providing alternatives to these closed systems, among other things.

Lasting Relationship

Legacy is not the only potential hurdle, the people create challenges of their own. Even clear mechanistic models, which do not suffer from the black box aspect of an AI model, still feel like a black box to employees who have no experience with them whatsoever. A lack of experience and comprehension leads to distrust. Timely involvement of the right stakeholders in a digital twin project is therefore paramount.

The importance of a flexible data platform becomes clear once again, as it gives the necessary room to maneuver in the development of a project.  In general, all attendees are looking for a balance between human knowledge on the one hand and the value derived from data and models on the other. The combination of IT, data, industry knowledge and academic expertise is not obvious, but very important. "This is exactly why Cognizant and CESPE are collaborating," stresses Sean Heshmat, GGM Data and AI Head at Cognizant.

The Value of a Digital Model

Attendees have very different views on the value of data and models. Some look at drug development as their only priority and digitization as little more than a means to an end. Intellectual property (IP) of data and models is not really considered.

Others are taking a more nuanced approach. They do see value in its digital twin models and the IP associated with them.

Niche Applications, Broad Challenges

During the roundtable, presentations and discussions seem to focus on niche applications of digitization and digital twins in the pharmaceutical sector. This makes sense: Cognizant and CESPE deliberately put a lot of effort into supporting the life sciences. However, a slightly more eagle eyed view reveals that the sector is facing the same problems and questions as many other industries. Chemical and manufacturing certainly come to mind.

Ultimately, everyone is looking for ways to digitally simulate complex processes, to gain deeper insights and find greater efficiency. An efficient and scalable approach that takes into account compliance requirements is essential for this endeavor. The TwinOps platform shines in this regard: the data layer is the sought-after foundation that enables rapid experimentation.

Puzzling Time

All the pieces of the puzzle are there. The real challenge now is to get started. There may be few first movers in the industry but there is no shortage of fast followers. prof Christoph Portier, CESPE Manager concludes, "In order to fully harness the potential of digital twins and drive innovation at an accelerated pace, it is crucial for companies in the chemical, pharmaceutical and biopharmaceutical industry to collaborate and set up precompetitive projects with multiple partners. By pooling resources, knowledge, and expertise, we can collectively explore the possibilities of digital twin technology and create a foundation for sustainable growth and competitive advantage."