Image
Bridging the Translational Gap: Programmable Virtual Humans Offer a New Path for Drug Discovery
A team of researchers led by You Wu and Lei Xie of Northeastern University, with co-author Philip E. Bourne, dean of the University of Virginia School of Data Science, has proposed a bold new model for drug discovery — one that could make the costly, failure-prone process faster, safer, and more predictive of human outcomes. Their study, “AI-powered Programmable Virtual Humans Toward Human Physiologically-Based Drug Discovery,” published in Drug Discovery Today, introduces a new concept: programmable virtual humans. These richly detailed computational models would allow scientists to simulate how new drug compounds behave inside the body before a single human trial begins.
Closing the Gap Between Promise and Reality
Despite remarkable advances in artificial intelligence and molecular biology, most drug candidates that show promise in early laboratory tests fail in clinical trials. This “translational gap” has long been one of the biggest obstacles in pharmaceutical development.
“Right now, our experimental models — whether cell lines, organoids, or animal studies — only approximate human biology,” said Bourne. “The goal is to bridge that gap with a dynamic, data-driven model that behaves like a real human system. If we can simulate physiological responses accurately enough, we can eliminate many of the dead ends that slow the path to new treatments.”
From Molecules to Systems
Unlike existing digital twins used primarily in later-stage clinical modeling, programmable virtual humans would begin at the molecular level by integrating physics-based models of physiology, biological and clinical knowledge graphs, and machine learning models trained on massive omics datasets.
In this hybrid framework, researchers could “program” a virtual human with a candidate compound and observe predicted effects across scales, from molecular interactions to organ function. By capturing how drugs interact within complex biological networks, the models could forecast efficacy, side effects, pharmacokinetics, and toxicity earlier than ever before.
“This is about more than accelerating drug pipelines,” Bourne said. “It’s about reimagining how we model the human body as a living system, one that’s computationally accessible and experimentally testable. That could fundamentally reshape our approach to everything from Alzheimer’s to autoimmune disease.”
A Framework for Collaboration
The authors emphasize that realizing programmable virtual humans will require global, interdisciplinary cooperation. Data integration across scales, standardized model validation, and regulatory trust are among the key challenges ahead.
Still, the societal potential is enormous. A mature programmable virtual human framework could sharply reduce reliance on animal testing, prioritize only the most promising therapies for human trials, and accelerate precision medicine by tailoring simulations to individual patients.
The foundation for this work was shaped by Bourne’s vision to apply data science and computational modeling to improve how new medicines are discovered. Xie has advanced that vision over many years through pioneering research in artificial intelligence, systems biology, and pharmacology. Together, their collaboration has bridged disciplines and institutions to make the concept of programmable virtual humans a tangible reality.
Looking Ahead
The project grew out of a longstanding collaboration between Bourne and Xie, who have worked together for many years on data-driven approaches to drug discovery. Together with first author You Wu, the team is charting a roadmap that combines artificial intelligence, systems biology, and mechanistic modeling to make programmable virtual humans a practical reality.
“Programmable virtual humans will give us a computational microscope into how drugs interact with the human body, helping scientists design safer, more effective therapies from the start,” said Wu, associate research scientist at Northeastern University.
As the authors conclude, developing trustworthy, interpretable, and generalizable virtual human models could mark a paradigm shift in how medicine is discovered. By merging mechanistic insight with machine intelligence, they envision a future in which researchers can test thousands of hypotheses safely, ethically, and virtually — before ever reaching the clinic.
