Researchers at the École Polytechnique Fédérale de Lausanne (EPFL) have developed an artificial intelligence foundation model called Virtual Tissues (VirTues) that maps the spatial organization of cells and proteins within tumor samples. The model was described in a study published in Nature by Charlotte Bunne's Artificial Intelligence in Molecular Medicine group.
A tumor contains not only cancer cells but also immune cells, blood vessels, and other stromal components whose spatial arrangement influences how the disease progresses and responds to treatment. Spatial proteomics technologies can measure dozens of proteins simultaneously to capture this organization, but the resulting high-dimensional datasets are difficult to interpret and compare across different studies, cancer types, and protein panels.
VirTues addresses this challenge by learning a shared representation of tissue biology from diverse spatial proteomics datasets. The model uses a novel Transformer architecture that can integrate data even when different sets of proteins are measured across studies. This allows any new tissue sample to be analyzed within the same framework without retraining the model from scratch.
The researchers assembled what they describe as the largest open dataset of spatial proteomics measurements to date, comprising more than 12,000 images from over 5,000 patients across 31 clinical cohorts. Using this atlas, VirTues can compare patient groups, identify spatial patterns that distinguish them, and discover spatial biomarkers associated with disease progression and treatment response.
Unlike conventional computational models designed for a single predefined task, VirTues follows the foundation model paradigm: it is trained broadly on tissue biology and can then be applied to many different analytical questions. The underlying model remains constant while the specific question changes, enabling rapid validation of discoveries across independent patient cohorts.
Oncologists involved in the work, including Andreas Wicki of the University of Zurich and Olivier Michielin of Geneva University Hospitals, emphasized that computational modeling is essential for making the exponentially growing volume of tissue data actionable in clinical settings. Michielin suggested that incorporating VirTues into precision oncology tumor boards should be a natural next step.
The model forms the tissue component of a larger five-year project, "Virtual Patient Labs: AI-Driven Simulation and Diagnostics for Precision Oncology," for which Bunne received the 2026 Lopez-Loreta prize. The project aims to build computational models of individual patients' biology by combining VirTues with routine pathology, genetic information, clinical data, and other records.
Bunne noted that placing a new patient's sample into the atlas provides broader biological and clinical context, but that the greater challenge lies in moving from description to predicting how a tissue will change under a given therapy. Testing whether such predictions can improve clinical decisions is the focus of the coming years.
AI model maps tumor tissue to improve cancer care
This is an independent summary. The complete reporting, supporting context and any primary documents remain with Medical Xpress.