On Virtual Twin® models in rare disease research, with Piet van der Graaf

Welcome to the Pharma Forum Podcast. This is web editor Nicole Raleigh, and today I have with me Piet van der Graaf, Senior VP at Certara, a company providing predictive simulation, data-driven modelling, and AI tailored for drug development. Welcome, Piet.
Thanks for having me, Nicole.
So today, we’ll be discussing the potential of digital twins in rare disease research and drug development. With over seven thousand rare diseases affecting an estimated thirty million people in the US alone, these conditions have long been underfunded and under-researched, leaving patients with limited options.
However, virtual trials are emerging as a breakthrough. Before we sink our teeth into our central topic, though, Piet, perhaps you could share with listeners a little bit about your background and the journey that led you to doing what you do today at Certara.
Yeah. Sure, Nicole. So I suppose the common theme of my journey, which has now been almost thirty-five years working in pharma, has been modelling and simulation. So as a master’s student almost thirty-five years ago, unlike all my fellow students, I wasn’t particularly keen to do a lab-based research project, and I managed to find a project which was called chemometrics in those days. So this was kind of applying data science to link properties of structures to pharmacological properties such as pharmacokinetics.
So that’s where I got really interested in modelling and simulation, data science, machine learning.
So I then moved to London to do my PhD with Sir James Black, and he taught me how to use modelling and simulation to understand interaction of ligands and biological systems.
So he called that analytical pharmacology, and that really taught me how this can be applied in a drug discovery setting. I then felt I would need to learn how to apply that to in vivo systems. So I went to Leiden University where I learned about pharmacokinetic–pharmacodynamic modelling, and then finally moved to Pfizer where I learned about pharmacometrics, which is the equivalent of chemometrics but applied to pharmaceutical research, set up the preclinical modelling and simulation group, and that evolved into the discipline that we’ll talk about later, quantitative systems pharmacology. After Pfizer, I set up a small company with a former Pfizer colleague, and we sold that to Certara in 2015.
And that’s where I am now heading up the industry’s largest group focusing on mechanistic modelling and simulation called quantitative systems pharmacology.
Thank you for that, Piet. Quite a journey, so we’re going to unpack some of that for our listeners, and as you say, to this main theme of quantitative systems pharmacology. So you’ve been seeing firsthand that virtual trials offer a cost-effective and efficient way to collect data and drive clinical trial innovation. So I was wondering if you could explain for our listeners the growing role of digital twins in virtual trials in speeding up drug approvals generally to sort of set the scene.
Yeah. Absolutely. So maybe first, let me start by explaining what is a digital or virtual twin. So it’s really an in silico model on the computer of a physical object or a person or a process, and we use the digital twin to simulate its behavior to better understand how it works in real life. Now, importantly, the digital twin needs to be able to interact with its environment, read data, update the model. So it’s a very dynamic kind of model, of course. Now, they’ve been around for quite some time in a variety of industries like automotive, aerospace, defense, retail, consumer goods.
And actually, most people will have several digital twins on their mobile phones, such as a map linked to GPS and real-time traffic data to optimize your car journey or a weather app.
Now digital twins are also increasingly used in health care—think, for example, about smartwatches with health and fitness trackers.
Now in recent years, digital twins have also entered pharmaceutical R&D, and they are increasingly being used in drug discovery and development. Now in the context of pharma R&D, a digital twin is a virtual model of an actual patient that mirrors their biological, genetic, clinical, pharmacological behaviors.
But the main advantage of using digital twins in drug development is that they allow for simulations of virtual trials where we can explore questions about a therapeutic before involving real patients.
And when I say virtual trial, this could be an N-of-1 trial in a single digital twin representing one individual, or a larger trial of many digital twins and virtual patients.
Now the use of digital twins can shorten the timelines of clinical trials due to better designs, reduced costs because fewer trial iterations are needed, and increased probability of success because optimal doses, dose regimens, combinations, and patient subgroups can be identified ahead of clinical testing. So ultimately, the impact of digital twin technology is that every patient will receive the right medicine at the right dose.
Thank you for that, Piet. I mean, it’s fascinating science and, as you say, impactful for that personalization. So if we go into finer detail now, I believe you’re also the co-author of a study on Pompe disease that highlights how these digital twins can transform clinical trials. So if we take what you’ve just been describing for us on digital twins, how do digital twins simulate real-world patient responses specifically and predict those treatment outcomes and reduce, as you’ve been saying, the need for large trials? I mean, we’ve gotten down to the individual level.
Yes. So as I briefly mentioned earlier, the actual technology underlying digital twins in drug development is called quantitative systems pharmacology, or QSP in brief. Now broadly speaking, QSP is the quantitative analysis of the dynamic interaction between drugs and biological systems, and it applies concepts of systems engineering, systems biology, and quantitative pharmacology to study and model complex biological systems through iterations of computational modelling on the one hand and experimentation and clinical trials on the other hand.
Now the outputs of QSP models are virtual patients, and they can be personalized, and then they become digital twins.
So, initially, a QSP model is developed based of prior knowledge. The prior knowledge can include anything ranging from fundamental understanding of the disease pathophysiology and the drug’s mechanism of action, pharmacology, pharmacokinetics, results from clinical trials with the actual drug or other drugs for the same disease, biomarkers, omics data, etc. From this initial QSP base model, we can then simulate the predicted response to a therapeutic for typical average virtual patients. However, because we know, of course, that not all patients respond in the same manner, the next step is to introduce biological variability into the model.
For example, assume that our drug target is an enzyme expressed in liver cells.
Then we can immediately think of already two factors that may influence the response: how many liver cells a patient has and how much enzyme is expressed in each cell.
Now if we assume for simplicity that an individual has an average, high, or low number of liver cells and similarly an average, high, or low number of enzymes expressed in each liver cell, we already have nine potential different patient phenotypes. Because a typical QSP model often contains dozens, if not hundreds, of biological variables, the number of possible virtual patients quickly grows to thousands, and their responses can span a wide range of clinical outcomes. However, we can calibrate that model with actual clinical data to ensure that the predicted outputs match the observed variability in real patients. Once we’ve achieved this, the final step is to match observed data for each individual patient with a unique profile generated by the QSP model—and now we have our digital twin.
Thank you for that, Piet. So it’s sort of self-explanatory at this point, the potential that this lends to rare-disease R&D. But I was just wondering, to make it crystal clear for listeners, when it comes to the role of virtual trials in collecting remote data to overcome small sample sizes in rare diseases, can you broaden that out for us?
Yeah. Sure. A digital twin is just one virtual patient in a virtual-trial output. So we can run virtual trials with hundreds, if not thousands, of virtual patients. We can mimic the design of an actual planned clinical trial exactly to match how it will run in real life. The advantage is that we can simulate many virtual trials to explore optimal design, optimal dose regimens, and more before running the actual trial. It gives us the opportunity to explore the best approach before investing in what is typically a costly and lengthy real-world trial.
Absolutely. Okay, so let’s segue now and put on our regulatory hats, if you will. How are regulators incorporating AI and machine learning to streamline the approval process, ensuring faster and safer drug development?
Broadly speaking, regulators have embraced modelling and simulation—including QSP, AI, digital twins, and virtual trials—heavily in recent years.
For example, there was a recent paper by FDA showing that the number of regulatory submissions supported by QSP models has grown exponentially in recent decades. And I was just before this meeting, Nicole, reading the annual report of the FDA’s Office of Clinical Pharmacology, where they summarize the impact of QSP on regulatory decision-making. And I cite from the report:
“QSP mechanistically and quantitatively integrates biological, drug, and clinical-trial information to model and simulate drug response and variability in patient responses to inform drug-development decisions. QSP staff played a key role in reviewing QSP models, gaining critical information”—and here comes the important bit—“on dosing in patients who were not able to be studied in clinical trials and the safest way to address missed doses.”
So here they are really talking about digital twins in trials where, for whatever reason, some patients may have missed a dose or were not dosed at all. These mechanistic models can help substitute missing data.
Thank you for sharing that, Piet. And it goes back to that well-known idea: the poison is in the dose.
Yeah. Absolutely.
Okay. My final question for today. What do you believe the future horizon of AI employment in R&D will look like in terms of the virtual trials and digital twins of tomorrow?
There are many things I can think of, but let me highlight a few. First, an area I’m very excited about is synthetic controls, highlighted in a recent paper in Lancet Digital Health. The idea is: can we use models, virtual trials, and digital twins to substitute placebo groups in clinical trials? This is especially relevant in pediatric rare diseases, where patient numbers are extremely small and it is ethically undesirable to place children on a placebo arm. Digital twins may be the solution.
Second, I believe digital twins will further enhance diversity and equity in drug development. Because we can create digital twins of anyone, we can better fulfil our mission of finding the right medicine and dose for every individual.
Lastly, a very exciting new topic is combining digital twins with organ-on-chip technology. This allows the creation of what we call virtual triplets: the actual patient, their digital twin, and an organ-on-chip version of the patient in an experimental setting. That triangle will be incredibly powerful in the coming years.
Exciting indeed. Thank you for sharing, Piet. It’s been an absolute pleasure.
Thank you.
And that concludes another episode of the Pharma Forum Podcast. You can find more information about this episode, including a download link and information about previous instalments, at pharmaforum.com/podcasts.
The Pharma Forum Podcast is also available on Apple Podcasts, Spotify, Overcast, Pocket Casts, Podbean, and pretty much wherever else you download your podcasts. Just search for “PharmaForum” and hit subscribe. Don’t forget to visit our website where you can sign up for daily and weekly news and analysis bulletins. And follow us on BlueSky and LinkedIn at @PharmaForum. That’s all for now. Thank you for listening.