Virtual Tumor Model Predicts Response to Liver Cancer Immunotherapy

A research team at Johns Hopkins has built a computational framework to forecast which people with hepatocellular carcinoma (HCC) may benefit from a combination of immunotherapy and a targeted drug. The approach merges quantitative systems pharmacology with a spatial agent-based model, allowing scientists to simulate not just tumor size, but how tumor cells interact with surrounding tissues over time.

In the study, the platform was expanded to include fibroblasts, cells previously linked to immunotherapy resistance in liver cancer. Using a machine-learning calibration workflow, the model was tuned to data from real clinical trials, producing “virtual patients” whose predicted outcomes can be compared with observed responses.

The key advance is spatial modeling. By tracking where individual cell populations reside, the researchers map the tumor microenvironment architecture that governs treatment sensitivity. This includes the distribution and organization of fibroblasts, immune T cells, and cancer cells, revealing how physical and biochemical barriers can shape drug effectiveness.

The team tested the system against therapy scenarios involving cabozantinib, a targeted therapy that blocks tumor growth signaling, and nivolumab, an immunotherapy. Simulated response rates aligned closely with those reported in clinical trials, suggesting the virtual cohort behaves like real patients.

To validate biological realism, the researchers compared predicted tumor architectures with post-treatment tissue patterns. They also contrasted the microenvironments of responders and non-responders, identifying mechanisms tied to therapeutic failure.

A striking finding emerged in predicted non-responders: fibroblasts can remodel the microenvironment into an immunosuppressive structure. In simulations, fibroblasts assemble into a physical “wall” that prevents T cells from reaching tumor regions, even when immune cells are nearby.

Because architectural features are visible before therapy begins, the model could help stratify patients—flagging who is likely to respond and who might need alternative strategies. The framework is intended for future clinical validation, rather than immediate use in treatment decisions.

The work was supported by multiple funding sources including the National Institutes of Health and other agencies. Results were published online July 14 in the Proceedings of the National Academy of Sciences.