For more than a decade, computational scientist Juan R. Perilla of the University of Delaware has been working to digitally reconstruct a very particular structure of the human immunodeficiency virus (HIV). Perilla and his colleagues set out to create an active three-dimensional digital model of the virus shell, or capsid, that researchers could study and probe as if they were working with an actual particle. The processing power required to build the simulation was significant, according to Perilla, because the model needed to track how a change in one area would impact the interactions of all two million atoms in the particle.
Perilla and his group succeeded in constructing the model and demonstrating various means of testing the simulation to ensure it behaves as it would in the real world. “You can actually interrogate the simulated particle, pushing and pulling on the capsid as if you were testing the actual physical system,” Perilla says. “You forget that it’s a digital copy that has been validated physically.”
The work has yielded at least one clinically useful discovery, revealing that the capsid shell is not rigid as scientists believed previously, but actually can be deformed, allowing the viral particle to slip through smaller-than-expected spaces. In a larger sense, Perilla’s research reflects the increasing interest in and progress toward designing biomedical digital twins, simulated models of biological phenomena and systems at multiple scales, from viral particles up to diseases, organs, and even entire people.
The concept of the digital twin first gained traction in industry, and although the potential applications have broadened, the basic idea remains the same. Generally, a working digital twin consists of three elements, according to computational biologist and complexity researcher James Glazier of Indiana University Bloomington. There is a near-real-time data feed reporting the evolving state of the original biological entity, machine, or part; a simulated model or representation of the time evolution of the original; and what Glazier calls a comparator, or a way of matching a predicted outcome to the observed outcome.
In manufacturing, engineers build a simulated model of an aircraft engine, based on its precise design specifications, that runs like the real-world version. They then outfit actual, operational engines with sensors that measure different variables in real time. As the sensor-equipped engine runs, the observed data is compared to the predictions of the simulated model to monitor whether or not the original is operating properly. “If there’s a divergence between the forecast and the observed value, that can indicate that there’s a problem with an engine, so you pull it out of service before there’s a failure,” Glazier explains.
General Electric and other companies have successfully used digital twins to ensure the safe and efficient operation of a wide variety of parts and products. Today, researchers are constructing digital twins of forests, warehouses, cities, and even planets. Applying these systems to biology and modeling the workings of organs or even entire individuals is a much more difficult job but one that is garnering attention. The National Academies of Sciences, Engineering, and Medicine recently established a committee to outline the major research needs and potential directions for the nascent field.
Expanding from industry to bio-medicine is no small leap, according to experts. “The difference is that you know how that engine works,” notes computational biologist and practicing surgeon Gary An at the University of Vermont. “You have a precise mechanical understanding of how that system behaves. The challenge with biology is that there is perpetual epistemic uncertainty with regards to how correct our specifications happen to be.”
Biological systems do not follow the script of a strict design document or set of plans, either. “Biology is about interactions, it’s about dynamics, it’s about self-organization and emergent properties,” adds Glazier. “And so just knowing the parts lists and static snapshots of the state will not tell you what is going to happen next.”
Yet the potential benefits of bio-medical digital twins are enormous. If doctors had access to a realistic digital twin of a hospital patient, then instead of prescribing a treatment and waiting weeks or months to see how it works, they could shorten the feedback loop, consistently measuring the observed impact against the intended outcomes. This responsiveness would be enormously helpful in fighting rapid onset conditions like sepsis, the cascade of reactions to an infection that kills at least 375,000 American adults each year.
Reinhard Laubenbacher, director of the Laboratory for Systems Medicine at the University of Florida, whose efforts to build a digital twin of the human immune system were profiled in a previous Communications article (https://bit.ly/40EOqNf), is working to construct a simulated model of pneumonia. Today, patients with severe pneumonia are treated in a hospital’s intensive care unit. Machine learning algorithms can predict, with extremely high accuracy, whether those patients will survive that infection. What the algorithms do not provide are possible interventions that might increase the patient’s chance of survival. This is the job of the physician, of course, but a digital twin could assist the doctor in this effort.
Laubenbacher and his colleagues are hoping to build simulations that model the patient’s immune response to the infection that led to the pneumonia. This would allow doctors to quickly evaluate different treatments in simulation and, ideally, help them choose one with the highest likelihood of changing the patient’s risk assessment for the better. The twin would augment the physician’s capabilities.
The complexity of the model and the hardware needed to run it present significant challenges. Perilla needed supercomputers to create his digital representation of the HIV capsid. If doctors are going to operate a digital twin in real time, they probably will not be able to run complicated multiscale models, but will have to rely instead on simpler, abstract versions that demand less computational power. Researchers would need to build models that can run on a handheld tablet and still make accurate predictions—no easy task.
Modified variations of digital twins are currently being used as decision support tools for cardiac patients. The company Heartflow developed a technology that uses a computed tomography (CT) scan of an individual’s heart, along with computational fluid dynamics (CFD), to build a three-dimensional representation of the heart which simulates blood flow through the arteries. This digital twin is more of a snapshot than a living, updating model, but it can highlight potential problems in blood flow. If bypass surgery is required, the surgeon can then test different placements of the blood vessel to be surgically implanted and see how each affects blood flow in the CFD simulation. Laubenbacher notes that this doesn’t exactly fit the paradigm of the digital twin, since the model of the patient’s heart is not evolving in response to real-time data, but it is a real-world simulation that is improving outcomes today.
“The challenge with biology is that there is perpetual epistemic uncertainty with regards to how correct our specifications happen to be.”
Similarly, Perilla’s highly advanced and realistic model might not technically be classified as a digital twin, since it is not updating based on real-time data collected at the source, but his early work showing the capsid’s unexpected properties in simulation also had a significant impact, inspiring a new form of drug treatment.
“Ten years ago, it was even harder to convince people that these simulations were worth doing,” Perilla recalls. “We’ve been swimming against the stream for a long time. But it’s validating now to see that the systems you’ve been working on have a real chance to improve people’s quality of life in these communities.”
As part of an information-gathering meeting to inform its new initiative, the National Academies convened a wide range of experts on biomedical twins, including Perilla, Laubenbacher, Glazier, An, and others, to outline the needs and opportunities ahead, including the variations of digital twins that could be useful, the data required, and the tools needed to collect this data and use it to inform the simulation. Some of this hardware exists today, according to Laubenbacher, while some does not; he expects the efforts and attention of the National Academies will help frame development.
His work on pneumonia, for example, would benefit from technology that allows doctors to gather detailed data on what is happening inside a patient’s lungs. “You need to be able to get measurements off that patient that you might not get from a blood draw, for example, and they need to be largely non-invasive because the patients are extremely sick,” explains Laubenbacher.
Perilla is excited about the technological aspects of this challenge and the possibility of building real-time, consistently updating digital twins of previously out-of-reach systems. “Larger scales have had an opportunity to advance more rapidly compared to smaller scales, but it certainly creates an aspirational perspective for work,” says Perilla. “We would indeed like to build models that can receive feedback in real time. Developing new technologies to be able to capture the behavior of the systems in an automated manner will be the future of the field.”
“Developing new technologies to be able to capture the behavior of the systems in an automated manner will be the future of the field.”
Despite the scale of the task, and the technology development to-do list, researchers believe there could be significant progress toward building more of these useful biomedical simulations in the years ahead. “There’s a sense of excitement, not that we could deliver the ultimate medical digital twins today,” Glazier says, “but that we can make that happen, and not in 20 or even 10 years, but in five years.”
Further Reading
Bryer, A., Rey, J., and Perilla, J.
Performance efficient macromolecular mechanics via sub-nanometer shape based coarse graining, Nature Communications 14, 2023.
Laubenbacher, R., Sluka, J., and Glazier, J.
Using digital twins in viral infection, Science, Vol. 371, No. 6534, 12 Mar 2021.
Laubenbacher, R. et al.
Building digital twins of the human immune system: toward a roadmap, npj Digital Medicine 5, 64 (2022).
Jones, D. et al.
Characterising the Digital Twin: A systematic literature review, CIRP Journal of Manufacturing Science and Technology, Volume 29, Part A, 2020.
Boulous, K. and Zhang, P.
Digital Twins: From Personalised Medicine to Precision Public Health, Journal of Personalized Medicine, 11, 745, 2021.
Join the Discussion (0)
Become a Member or Sign In to Post a Comment