At Johns Hopkins, researchers have built a computational mirror of liver cancer — a virtual tumor that thinks, resists, and responds much as a real one does. By simulating how individual cells behave in space, the model can predict which patients with hepatocellular carcinoma will benefit from a combination of immunotherapy and targeted therapy, and which will not. The discovery that fibroblasts form a literal physical wall between immune cells and their targets offers a new lens through which oncology might finally move from trial and error toward something closer to foresight.
Johns Hopkins develops virtual tumor model to predict liver cancer immunotherapy response
Even if immune cells were located near the tumor, the fibroblast would block them from reaching it.
Mark
Why does it matter that you can simulate a tumor instead of just testing drugs in the lab?
Mimi
Speed and scale. A phase I trial might have 15 patients. From that small group, the model can generate a virtual population the size of a phase III trial—thousands of patients—and predict how a therapy might perform in a much larger study without exposing anyone to risk.
Mark
So you're saying you can skip some of the human testing?
Mimi
Not skip it entirely. You still need to validate the model against real outcomes, which we did. But you can compress the exploration phase. Instead of trying three different drug combinations in sequence, you simulate them first and pick the most promising one.
Mark
The fibroblast barrier sounds like a specific discovery. Did you know that was going to be the problem?
Mimi
We suspected fibroblasts played a role in resistance, but the model showed us exactly how—they form a physical wall. That's the kind of detail you might miss in a petri dish or even in early clinical data. The spatial dimension matters.
Mark
Can you use this to predict who will respond before treatment starts?
Mimi
That's the goal. The architectural features the model identifies are visible in tumor biopsies before any therapy begins. If we can link those pre-treatment patterns to predicted response, we could eventually tell a patient upfront whether a particular treatment will work for them.
Mark
How close are you to using this in actual patient care?
Mimi
The model needs more validation first. We've shown it works in retrospective analysis—checking predictions against known outcomes. But before a doctor could use it to guide treatment decisions, we'd need prospective studies confirming it works in real time.
O Pulso
- Liver cancer progresses so rapidly that physicians often run out of time before finding the right treatment — the urgency of this reality drove researchers to seek a faster path.
- A virtual tumor platform developed at Johns Hopkins now simulates how real tumors and their surrounding cells respond to drug combinations, matching outcomes seen in actual clinical trials.
- Fibroblasts — connective tissue cells long suspected of shielding tumors — were confirmed to build physical barriers that block immune cells from reaching their targets, explaining a stubborn pattern of treatment failure.
- Because these fibroblast structures are visible in tumor tissue before treatment begins, they could one day serve as predictive markers, allowing doctors to choose the right therapy from the start rather than discovering failure months later.
- Clinical validation is still required before the model enters practice, but the foundation is laid for a shift from reactive experimentation to simulation-guided treatment selection.
At Johns Hopkins, researchers have built a computational mirror of liver cancer — a virtual tumor that thinks, resists, and responds much as a real one does. By simulating how individual cells behave in space, the model can predict which patients with hepatocellular carcinoma will benefit from a combination of immunotherapy and targeted therapy, and which will not. The discovery that fibroblasts form a literal physical wall between immune cells and their targets offers a new lens through which oncology might finally move from trial and error toward something closer to foresight.
A team at Johns Hopkins has created a virtual tumor that behaves like a real one — a computational model designed to predict which patients with hepatocellular carcinoma, a primary liver cancer, will respond to a combination of immunotherapy and targeted therapy. The work, published in the Proceedings of the National Academy of Sciences, was motivated by a simple and urgent problem: many cancers move too fast for doctors to experiment with treatments one by one.
The platform fuses two mathematical approaches — quantitative systems pharmacology, which models whole-body responses to treatment, and an agent-based model that tracks individual cells and their positions in space. Together, they map not just how many cells exist, but where they sit and how they interact. A machine-learning calibration system tunes the simulation against real clinical trial data, generating virtual patients whose predicted responses can be checked against actual outcomes.
When the team simulated treatment with cabozantinib and nivolumab — a targeted therapy and an immunotherapy drug — the virtual patients' response rates matched those from real trials. More revealing was what the model showed about treatment failure: fibroblasts, a type of connective tissue cell, were found to physically remodel the tumor microenvironment, forming a barrier that blocked immune cells from reaching their target even when those cells were nearby and activated.
The practical promise is significant. These fibroblast-heavy architectural patterns are visible in tumor tissue before any treatment begins, meaning a biopsy could one day reveal whether a patient's tumor is likely to resist immunotherapy — allowing a different approach to be chosen upfront. For cancers that leave little room for experimentation, the ability to simulate before treating could matter enormously.
A team at Johns Hopkins has built something that sounds like science fiction but works like a practical tool: a virtual tumor that behaves like a real one. Researchers from the Kimmel Cancer Center and the School of Medicine created a computational model capable of predicting which patients with hepatocellular carcinoma—a primary liver cancer—would respond best to a combination of immunotherapy and a targeted drug that starves tumors of growth signals. The work appeared online July 14 in the Proceedings of the National Academy of Sciences, supported in part by the National Institutes of Health.
The motivation is straightforward and urgent. Many cancers progress so quickly that doctors don't have time to experiment with different treatments or surgical approaches. Atul Deshpande, an assistant professor of oncology at Johns Hopkins and senior author of the study, describes the thinking this way: instead of trying one therapy, then another, then a third, what if you could simulate different doses and combinations in silico first? The model could point physicians toward the most promising path for each patient. The tool still needs further validation before it enters clinical practice, Deshpande notes, but the foundation is there.
The virtual tumor platform combines two mathematical approaches. The first is quantitative systems pharmacology, or QSP—a set of equations that models how the whole body responds to treatment, including tumor progression and drug effects. The second is an agent-based model that tracks individual cells and their behavior. Together, they create a map not just of how many cells exist but where they sit in space, allowing researchers to predict how the tumor and its surrounding microenvironment will evolve. Aleksander Popel, a professor of biomedical engineering and oncology at Johns Hopkins who developed the platform, led this expansion to include fibroblasts, a cell type long suspected of helping tumors resist immunotherapy. The team also built a machine-learning calibration system that tunes the simulation to match data from actual clinical trials, generating virtual patients whose predicted responses can be tested against real outcomes.
When the researchers simulated treatment with cabozantinib, a targeted therapy, and nivolumab, an immunotherapy drug, the virtual patients' response rates matched those reported in real clinical trials. This validation gave the team confidence that their digital tumors behaved like actual ones. They also compared the model's predicted tumor structures against real tissue samples taken after treatment and examined the microenvironments of patients who responded versus those who didn't.
What emerged was striking: fibroblasts, a type of connective tissue cell, remodeled the tumor microenvironment in ways that suppressed immune function. Among virtual patients who failed to respond to therapy, the model revealed that fibroblasts formed a physical barrier. Even when immune cells were positioned near the tumor, these fibroblasts blocked them from making contact. It's a literal wall between the immune system and its target—a mechanism that explains, at least in part, why some patients don't benefit from immunotherapy despite having activated immune cells.
The practical implication is tantalizing. If multiple treatment options exist for a given cancer, the model could help determine which would work best or which to avoid entirely. Because the architectural features the model identifies—the fibroblast barrier and other structural patterns—are visible in patient tumors before any treatment begins, they could eventually serve as predictive markers. A biopsy or imaging study before therapy starts might reveal whether a patient's tumor has the kind of fibroblast-heavy architecture that would resist immunotherapy, allowing doctors to choose a different approach upfront rather than discovering the problem months into treatment.
The study was co-supervised by Popel and Elana Fertig, director of the Institute of Genome Science at the University of Maryland School of Medicine. The work represents a shift in how oncology might approach the problem of treatment selection—not through trial and error, but through simulation. For cancers that move fast and leave little room for experimentation, that difference could matter enormously.
Citações Notáveis
Our idea was to create a computational model where we could simulate trying different doses or combinations of cancer therapies, and it could help guide physicians toward the best options for patients.— Atul Deshpande, assistant professor of oncology at Johns Hopkins
Quer a matéria completa?
Leia o original em Medical Xpress ↗
Health & Medicine