Jon Walsh, founder, chief scientific officer, Unlearn, explains how regulators are clarifying best practices for integrating AI and digital twins into clinical research.
In a recent interview with Applied Clinical Trials, Jon Walsh, founder and chief scientific officer of Unlearn, discussed how AI-designed therapies and digital twin technologies are reshaping the clinical trial landscape. While the fundamental principles of clinical testing remain consistent, Walsh explained that AI-discovered drugs often warrant deeper attention to specific biomarkers and safety signals in early development. He also highlighted how digital twins—virtual models of trial participants—can significantly improve trial efficiency, reduce control group sizes, and enhance the precision of treatment effect estimates. As regulatory agencies like the FDA and EMA begin to offer clear frameworks for integrating AI into drug development, Walsh emphasized the importance of transparency, reproducibility, and interpretability in model design and implementation.
ACT: What are the regulatory implications of using AI-generated data or digital twins in trials?
Walsh: Regulators have been very clear in their positions and guidance on how to use AI in drug development more safely, effectively, and in line with their expectations. For example, the FDA released guidance earlier this year outlining a seven-step process for applying AI in this context. Similarly, EMA has been developing frameworks, such as its 2024 reflection paper, which emphasize the importance of context of use.
Sponsors need to clearly specify how a model will be used in drug development, identify potential risks and impacts to the study, quantify those risks, and have a plan to address them by evaluating the model under defined conditions.
So far, the primary way we use digital twins in randomized controlled trials has been fully aligned with regulatory guidance. From the outset, we designed our approach to help randomized trials achieve higher statistical power, in a manner consistent with both EMA and FDA expectations. We have received positive feedback from both agencies—through EMA qualification programs and open discussions with the FDA, where we’ve even published their comments.
When it comes to other trial designs, such as open-label studies—where there is more risk and obtaining a control group is often not feasible—there are still clear ways to use digital twins within the FDA’s framework. These methods allow regulators to assess the risks and interpret evidence appropriately, even in these more challenging contexts.
Full Interview Summary: AI-designed therapies can present unique characteristics in clinical trials compared to traditionally developed treatments, often requiring closer scrutiny of biomarker signals and safety data early on. However, the fundamental process of clinical testing remains consistent. Digital twin technology helps make these trials more efficient and insightful by modeling individual trial participants under standard of care, offering a virtual comparator without needing additional enrollees.
Digital twins work by creating a probabilistic model of how a specific patient would respond to standard treatments over time. Rather than enrolling new participants, sponsors gain deeper insights into existing ones, improving the precision of treatment effect estimates and boosting trial power. In randomized controlled trials, this added statistical confidence can reduce participant numbers or accelerate timelines. In open-label studies, digital twins can serve as virtual control arms, allowing meaningful comparisons when a real control group is not feasible.
Regulatory agencies like the FDA and EMA are providing clear guidance on the use of AI and digital twins in drug development. They emphasize defining the model’s context of use, evaluating its risks, and proving its reliability. The companies building digital twins must ensure model development is transparent, reproducible, and interpretable—tracking how models were built, and which data were used, while safeguarding against unblinding.
Looking ahead, digital twins are expected to become a foundational part of the clinical trial landscape, particularly in neuroscience and CNS trials. They offer the promise of more adaptive, efficient, and patient-centric designs that can accelerate drug development and reduce reliance on traditional control groups, while maintaining—or even improving—the quality and reliability of clinical evidence. As adoption grows, digital twins could reshape how sponsors design trials and regulators evaluate new therapies.
