The data science and cancer research communities are abuzz with the talk of “digital twins”—a real-time, virtual representation of a living physical system. Following the release of the National Academies of Sciences, Engineering, and Medicine (NASEM) report, which included biomedical applications, efforts around digital twins in medicine have accelerated quickly. In a 2024 review, the authors note that the number of publications featuring digital twins in healthcare went from zero in 2016 to more than 80 in 2023. The fact that NASEM recognized cancer as the exemplar biomedical digital twin for their report further underscores this technology’s tremendous potential in cancer as well as cancer’s impact on advancing digital twin technology.

Recognizing What’s Possible in Digital Twin Technology

You’ve likely heard a lot of “noise” around the digital twin topic, ranging from strong skepticism to passionate advocacy. Whether you’re interested in this technology from a research or a clinical perspective, it’s hard to know what’s truly feasible or why you should even care. The technology seems so far in the future, is it even applicable today?

When considering these questions, it’s important to keep in mind that we’re already familiar with many of the elements that form the foundation for cancer digital twins.

Although these current models have limits, they serve as key starting points for improving both insight and decisions.

What is new, and what’s really driving digital twins, is how we’re using data and models to make decisions on a different level. Using technology, we can examine cells within a specific tumor, a specific cancer, even a specific patient. Not only that, we can also monitor and incorporate so much more information about the patient’s history and even how they respond to treatments, both at the tumor and whole-body level. As we’re able to merge data, artificial intelligence, and mechanistic models, we’ll continue to make progress toward realizing the goal of precision oncology, enabling us to give each patient the best care possible over the course of a lifetime.

Embracing a Paradigm Shift for Model-Enabled Science

Developing a truly translational and dynamic digital twin, however, challenges the very foundation of many existing predictive cancer models. For example, we often begin with a research question, such as, “Will this treatment kill the cancer while not harming the patient?” Then, we adjust our models to best fit our observations across a range of patients, either because we don’t have enough data, our question is too complex, or our results have too much uncertainty. We’re left with a corrected model. That model’s fine for addressing one question across a range of patients but loses accuracy when we apply it to individual situations that change over time—such as over the course of cancer or over a lifetime.

For digital twin technology to flourish, we need to envision our models in a new way, one that enables those models to be both personalized and suitable for use over time.

It may seem daunting (and even impossible), particularly when you realize the fact that true digital twins, at least by the NASEM definition, are not just a one-time decision. Instead, true digital twins involve an ongoing series of model updates, insights, and decisions—all tied to a specific system or object, such as a patient, a tumor, or even a cell.

Recognizing the limitations of our existing models is a stepping stone on the path to realizing digital twin applications in cancer. By capitalizing on team science to make models more widely usable, we’ll be able to reach our digital twin destination.

Supporting Patient-Centered Collaborations

Fortunately, many collaborative efforts are well underway, as NCI’s divisions, offices, and centers are all working to produce new data and models for cancer research.

For example, NCI’s:

• Cancer Research Data Commons serves as a nexus for data and resources for building models.
• Office of Data Sharing is leading efforts to make those data easier to find, use, and share.
• CBIIT, as part of the NCI-Department of Energy (DOE) collaboration, recently funded efforts for digital twins in radiation oncology, in collaboration with the Division of Cancer Treatment and Diagnosis.
• Frederick National Laboratory for Cancer Research is establishing partnerships for pursuing digital twins. Some of that work includes supporting the NCI-DOE collaboration, the RAS initiative, and the national Cryo-Electron Microscopy facility for cancer.

With so much existing support in the pursuit of cancer digital twins, the only question remaining is: “When will we start to see these twins come to fruition?”

Based on the NASEM report, a recent report from the First Virtual Human Global Summit, several early projects for cancer digital twins, and insights from a recent NIH Interagency Modeling and Analysis Group meeting, we now have some clear guidelines for developing and using twins. Here are three key areas we need to keep in mind as we start:

• Address a specific need.
  If we break a digital twin into manageable parts and have enough information to put the pieces together, we can use a team science approach to handle the complexity of the various models and mechanisms and, ultimately, bring everything together.
  Conceptualizing a fully complete and very complex cancer digital twin (with the ability to answer every potential question) may be captivating, but it’s not very practical. Instead, we need to focus on addressing a specific need, answering a specific question, or supporting a specific decision and use the information that’s available to produce meaningful results. If we break a digital twin into manageable parts and have enough information to put the pieces together, we can use a team science approach to handle the complexity of the various models and mechanisms and, ultimately, bring everything together.
• Set expectations appropriately. By developing early cancer digital twins with existing models (which have their limits), we will, in turn, help make these individual models better at making predictions in specific cases. Some of these models will be easier to incorporate in a digital twin than others, depending on the available information, complexity, and question at hand. We also will need new and greater amounts of data to support composing cancer digital twins. Fortunately, each model we develop for a specific cancer digital twin has the potential to advance (or serve as a basis) for future cancer digital twins.
• Work as a team and as a community. In making a successful cancer digital twin, particularly one that aims to involve a specific individual, we need to keep the patient at the center. We’ll also need a concerted effort to educate and engage both the community as a whole and the team developing the digital twin. This includes the person we’re “twinning,” as well as clinicians, researchers, data scientists, and many others. By working together, we can share insights and identify gaps in the information we need to develop, validate, and deploy digital twins in health.

Taking the Next Steps Toward Cancer Digital Twins

There is a clear path forward for cancer digital twins. The next steps will be to continue to broaden our perspective. By viewing the mechanisms and drivers underlying cancer (and the impact on the individual) through a computational lens, we can extend our existing hypothesis-driven research perspective. Fortunately, we already have many models to draw from, which will help address individual pieces of an increasingly complete digital twin. With further collaborations and open sharing of information, today’s digital twin approaches will be part of tomorrow’s routine care.