March 7, 2023 PAO-03-23-NI-02
The concept of a digital twin is somewhat self-explanatory: it is a virtual, digital copy. That copy can be of many things, though — a physical object, such as a product; a process, such as a manufacturing process; a system, such as a manufacturing facility; a network, such as a supply chain; or infrastructure, such as a city.1,2 Digital twins are also more than digital copies; they are connected to their real-world counterparts from which real-time data (e.g., enterprise, Internet-of-Things (IoT)) are gathered and generated results are returned. The real world is simulated in the virtual environment using these data to predict possible performance outcomes and evaluate what-if scenarios. Using digital twins can help companies better understand their products, processes, and systems to facilitate better decision making and enable avoidance of future problems.
NASA developed the first digital twins in the 1960s, making copies of its spacecraft to perform simulations and train astronauts.1 These digital twins played a key role in saving the lives of the astronauts on the problem-plagued Apollo 13 mission. By the 1970s, mainframe computers were used to support basic digital twins of large facilities. In the 1980s, 2D computer-aided design (CAD) systems were used to provide technical drawings digitally and further build out initial digital twin solutions. The next decade brought 3D CAD with parametric modeling and simulation, and the following decade saw these systems cloud-connected for collaboration and program management.
Today, real-time 3D digital twins are developed using not only many different physical models and the IoT but artificial intelligence (AI), machine learning (ML), natural language processing, the “metaverse,” and virtual and augmented reality (VR/AR).3 For companies with the appropriate level of digital maturity, strong data infrastructures, experts with the right skills, and the ability and willingness to invest upfront in this new technology, they are improving performance in many industries and various applications.2 Not only have improvements in digital technologies had an impact; automation systems and real-time analytical tools are providing the reliable and robust data needed to maximize the use of digital twins.
Digital twins provide an in silico means for optimizing products, processes, and systems at a much lower cost than what is required to perform physical experiments.1–3 They also help reduce waste by predicting optimum maintenance scheduled and preventing problems. They are being used for many applications,3 including the modeling of Los Angeles’ transportation infrastructure and the entire city of Shanghai, China. A digital twin of Sofi Stadium in Los Angeles models the stadium and 300 surrounding acres. Tesla creates digital twins of all cars it produces to minimize the service and maintenance required.
Within the pharmaceutical industry, numerous potential applications for digital twins exist.4,5 Perhaps one of the most obvious is the use of digital twins of manufacturing processes to increase efficiency and productivity by facilitating process optimization and scalability, as well as enabling predictive maintenance. There are less obvious uses as well. Digital twins of cells are determining if surgery is needed in heart patients and accelerating drug discovery, while drug development could be improved from a cost and time perspective through the use of digital twins as replacements for the placebo-control arms of clinical trials. Digital twins of diseases could help researchers better understand disease mechanisms and the impacts of drugs on disease progression, while digital twins of organs, genomes, and patients could better enable precision/personalized medicines. Digital twins of hospitals and clinics, meanwhile, could potentially improve patient care.
Access to many types of omics data and sequencing information at the single-cell level combined with ML, particularly deep learning, and AI is enabling the development of digital twins of human cells for use in drug discovery.6 DeepLife is one company with a cellular digital twin platform based on single-cell omics data that it hopes will accelerate and optimize target identification. The cloud-based system can rapidly evaluate how different cells (up to 10 million) respond to billions of drugs and drug combinations, viruses, and other perturbations. In addition to discovery of novel drugs, it is also finding use in the repurposing/repositioning of existing drug substances/products.
Clinical trials account for the largest portion of drug development costs and generally are inefficient. They typically also do not represent the true patient population or real-life experiences of patients. The use of placebos also means that a portion of patients is denied treatment. Digital twins can help overcome these issues by simulating diverse patient characteristics, accelerating patient selection, and predicting patient responses to allow elimination of placebo control arms.7 They can also reduce the number of actual patients subjected to rea-world testing, increasing the efficiency, effectiveness, and safety of clinical trials while reducing their cost. That would enable more companies to run more trials and bring more needed medicines to market.
Another approach to addressing drug development challenges is the development of digital twins of human patients, organs, or cell receptors. Metaverse technologies are being deployed to provide immersive or AR visual representations of the effects of proposed therapies on these digital twins.8
A huge challenge in the pharmaceutical industry is the development of optimum robust and scalable processes from the outset. Doing so is imperative, however, to ensure cost-effective and reliable production of drug substances and drug products. Furthermore, making changes to approved processes involves the performance of extensive bridging studies — a lengthy and costly enterprise.
Digital twins have the potential to aid in the development of optimum processes that are robust, scalable, and commercializable.9 Most pharma companies already use computational fluid dynamics (CFD) software to model momentum, energy, and mass transport within reactors and bioreactors. Digital twins go a step further, incorporating CFD data with other information to predict optimal reactor designs, process parameters, and addition protocols. They can also be used to predict performance, thereby reducing the number of process performance qualification (PPQ) runs needed to establish robust control strategies, which can save millions of dollars.
Other benefits include more efficiently tracking processes to ensure compliance with regulatory requirements, improving data integrity, facilitation of the move to continuous processing, and reduced infrastructure costs.10 Digital twins can also be used to track maintenance and operational issues to identify equipment that must be repaired or replaced before it fails and reduce the need for skilled labor, which is in short supply within the entire biopharma industry today.11 Modeling using digital twins overall provides greater process understanding, which enables the development of more effective process control strategies and ultimately more robust processes.12 McKinsey estimates the savings for pharma companies that leverage digital twins could experience increases in productivity of as much as 150–200%.11
Models used in digital twins of pharmaceutical manufacturing processes can be based on mathematical or stochastic models (or a combination of the two) that are integrated with geometric modelling techniques.13 Currently used scale-down, physical models do not provide higher levels of process understanding and therefore have limited use.12 Mechanistic models used to develop digital twins, on the other hand, include assumptions based on known principles as well as data, providing a more comprehensive description of the process and thus improved predictive power. Such holistic models, which generally leverage AI and real-time data, can help identify bottlenecks and lead to improvements across entire processes and plants.
Successful pharma manufacturing digital twins have both physical and virtual (ideally comprehensive, but typically simplified) components that communicate with one another with the goal of enabling the U.S. FDA’s vision to develop a maximally efficient, agile, flexible pharmaceutical manufacturing sector that reliably produces high-quality drugs without extensive regulatory oversight.14
The physical component serves as the source of data, including critical process parameters (CPPs) for the equipment and process and critical quality attributes of the product, values of which are monitored using process analytical technology (PAT).14
Virtual components perform real-time process simulations and system analyses, the results of which are sent to a data management platform for visualization and used to create control commands that are delivered to the physical process to ensure optimal control and performance.14 Modeling platforms include MATLAB and Simulink (MathWorks), COMSOL Multiphysics (COMSOL), gPROMS FormulatedProducts (Process Systems Enterprise/Siemens), aspenONE products (AspenTech), and STAR-CCM+ (Siemens), among others. Cloud-based, IoT, software-as-a-service data management platforms include Predix (General Electric), Mindsphere (Siemens), SEEQ, TrendMiner, TIBCO Cloud, and others.
To be maximally effective, digital twins require accurate data. In many pharmaceutical industry applications, those data may be sensitive, personal patient data — particularly for digital twins focused on supporting personalized medicines.4 Ensuring data security is therefore essential. Hesitancy surrounding AI applications and their use of personal data make this challenge more complicated.7 One possible solution is to leverage another advanced solution — blockchain technology — to secure data.4
Digital twins for the pharma industry must also be created using a hodge-podge of data sources ranging from genomics information to cell culture process parameters and physician notes, creating the need for solutions that can manage disparate data types and forms.7
Beyond data integrity issues, regulatory acceptance and the general willingness to adopt digital twins may be additional concerns.7 Successes are most likely to occur where digital twins represent continued advances in modeling, namely for pharma manufacturing. The concepts of digital twins for cells, organs, and patients and their use in clinical trials will be more slowly recognized.
Other larger issues relate to the upfront investment required to create digital twins and the time required before benefits can be realized.13 Firms seeking to implement digital twins must not only be prepared for the effort and commitment required, but also have sufficient digital maturity and access to large quantities of appropriate, reliable and robust data, such as through the use of state-of-the-art PAT solutions for upstream and downstream processing.
Despite these hurdles, pharma manufacturers are pursuing digital twins for enhancement of production capabilities. GlaxoSmithKline (GSK) and Sanofi are two examples.
GSK worked with French IT firm Atos and the engineering company Siemens to pilot a digital twin of one of its vaccine manufacturing processes, specifically a process for the production of a particular vaccine adjuvant.15 The digital twin is connected to the real-world manufacturing process, allowing physical sensors to send data to the twin and the twin to provide simulated insights to the control system of the physical process. GSK is able to fine-tune the process in real time using a range of models and ML techniques, and as importantly identify and eliminate or control manufacturing variabilities, reducing risk and facilitating tech transfer. Based on the successes achieved during the pilot, the company expects to leverage digital twins for other vaccine manufacturing processes, potentially as simulators for operator training and ultimately see them used as routine tools.
Sanofi, meanwhile, is also exploring the use of digital twins for vaccine manufacturing.16 In its case, the goal is to establish optimal processes before actual deployment. For its software partner, the company turned to Dassault Systèmes and its simulated 3D spaces, a platform that Sanofi is using to create virtual manufacturing systems that mimic processes Sanofi has under development.
David is Scientific Editor in Chief of the Pharma’s Almanac content enterprise, responsible for directing and generating industry, scientific and research-based content, including client-owned strategic content, in addition to serving as Scientific Research Director for That's Nice. Before joining That’s Nice, David served as a scientific editor for the multidisciplinary scientific journal Annals of the New York Academy of Sciences. He received a B.A. in Biology from New York University in 1999 and a Ph.D. in Genetics and Development from Columbia University in 2008.