Digital twins are rapidly gaining traction in the personalised medicine space – but can they make an impact in drug development too? TwinEdge Bioscience thinks so. Chief Business Officer Kevin Buyens spoke to us about how TwinEdge’s technology works, where it adds value, and why data-driven scale and individualised biological insight must go hand in hand.

TwinEdge Bioscience is working in the emerging field of digital twins – can you briefly define digital twins?

Honestly, I think the term ‘digital twins’ is often used too broadly, covering everything from simple data representations to highly complex simulations. But if I had to propose a definition in the field of medicine, I’d say that a digital twin is a computationally generated representation of a biological system (the patient), built from the data associated with that patient. It can be interrogated digitally, just like a tangible model can be physically challenged, unlike traditional bioinformatics or AI-based approaches, which are largely descriptive. In short, true digital twins must be dynamic and interactive – capable of answering ‘what if’ questions about disease drivers, treatment responses, resistance mechanisms and risk.

TwinEdge operates in this landscape with a focus on mechanistic biology. Rather than attempting to model the full complexity and variability of human biology and disease, we develop tissue-level models – particularly tumour tissue – which we prefer to call ‘digital avatars’.

So, how is TwinEdge positioned in the wider landscape?

While digital twins will undoubtedly play a central role in personalised medicine moving forwards, we feel we can have the greatest impact right now upstream, in drug development. Success rates here remain extremely low: only around 5–10% of oncology drugs that enter clinical trials are successfully approved. In most cases, we believe that these drugs don’t fail because they don’t work, but rather because the right patient population isn’t identified early enough. By enabling better clinical positioning, improved trial design and more precise patient selection, TwinEdge’s technology aims to increase these success rates – and even modest gains could translate into significant improvements in trial outcomes.

What is TwinEdge’s technology? What sort of data is it built from?

We have built a large, diverse population of over 20,000 digital avatars from high-quality multimodal data – primarily transcriptomic and genomic data, but we also incorporate proteomics, metabolomics, pathology and longitudinal clinical records where possible. Our key differentiator is how we combine AI and machine learning with mechanistic biology to produce predictions that capture the range of interactions and disease dynamics that shape outcomes. This is what makes our digital avatars an excellent system for testing possible drug responses and making informed decisions about future clinical trials.

How do you go about mitigating potential biases or blind spots?

As mentioned, to manage complexity while avoiding overreach, TwinEdge does not attempt to replicate the complexity of the human biological system, focusing instead at the tissue level. Rather than simply aggregating large volumes of data through statistical learning or generative AI models, we focus on constructing truly individualised digital avatars. Each avatar accurately represents the unique molecular make-up of a given patient’s tumour, supplemented by the patient’s clinical annotations.

In addition, it’s important to emphasise that our digital avatar population is not static. Not only do we incorporate new data types into the avatars as they become available – creating richer, more sophisticated digital patient representations – but we also continuously expand our collection. For example, when working on a specific cancer type, such as ovarian cancer, we enrich the dataset to better capture the mutational and molecular diversity of that cancer, as seen in clinical and research settings.

You’re talking about both individualised and population-level data: what exactly is the right scale for this sort of modelling?

Drug development is a population-level endeavour. But in practice, the dataset traditionally used to set a clinical strategy consists of a handful of preclinical models and perhaps a Phase I trial involving no more than 15–30 patients. These datapoints are rarely sufficient to predict how a drug will perform across large and diverse real-world patient groups. Only early access to a large population enables the mapping of different treatment responses, disease states and hidden subtypes. So scale is really important – and that’s exactly what our continuously expanding digital avatar population brings.

Moreover, what we do at TwinEdge is balance breadth with depth: we use small, programme-specific datasets to create corresponding patient avatars and then map them onto the broader population to provide clinical context. This allows researchers to see where their data sits within the wider clinical landscape and to explore how representative – or limited – their initial findings may be.

What’s more, because our models are mechanistic, we can push further, identifying patients within the larger population who share similar regulation mechanisms and biological drivers. This enables more accurate prediction of treatment outcomes and helps uncover opportunities that might otherwise be missed. For example, a therapy initially developed for a particular patient group can be assessed for potential effectiveness in other cancer subtypes with similar underlying mechanisms, supporting indication expansion and pivotal trial design.

We’ve heard a lot about the benefits of digital patient modelling, but is there a risk that increased reliance on these tools could standardise drug discovery and limit biological exploration?

It’s an interesting question. I think the answer hinges on how you view digital patient models: are they complementary tools within the existing drug development process or replacements for it?

At TwinEdge, we’re looking to augment and accelerate current workflows, not disrupt them outright. Our digital avatars sit within an iterative loop that still requires experimental and clinical validation. They can support more efficient drug development and avoid poorly designed trials through better upfront validation and scenario testing. But in the end, first preclinical and then clinical validation remains essential.

What’s exciting about our technology is that it helps uncover non-obvious biological mechanisms, novel combination strategies and previously unrecognised patient subgroups. When used thoughtfully, I believe it will expand rather than constrain scientific discovery by enabling researchers to think beyond their original assumptions, driving more informed, effective and innovative drug discovery and development.

How does your work align with other players in the digital twin space?

There are actually two other companies working in a closely related space at Biopôle: Calico Biosystems  and Prevision Medicine. Their technologies are positioned at a fascinating intersection of translational challenges, enabling drug testing in precious primary patient tissue. This aligns well with our work, as we can digitally predict which combination of drugs would most likely benefit that patient, and then the best options can be selected for physical testing on the primary tissue.

What are the next steps for TwinEdge – and what would you like to see in the future?

We’re at the start of an exciting journey. We envision a future where every participant in a clinical trial has a digital twin or avatar as a standard part of the protocol, assisting in the trial’s execution. This goal is perhaps not that far off: it should be achievable within the next three to five years. From there, it’s only a small leap to clinicians using our technology for ‘what if’ experiments to select personalised therapies for their patients. When quantum computing becomes a reality, our technology will facilitate the testing of all potential combinations of approved drugs to tailor treatment to specific patients.

But right now, our efforts centre on demonstrating our technology’s capabilities through real-life programmes with clients, particularly in areas like patient stratification, trial design and drug response predictions. We’re focused on demonstrating its potential as a digital asset generation and development engine that can be applied productively across the entire drug development spectrum.