Researchers have developed a new artificial intelligence foundation model capable of analyzing spatial proteome data to reconstruct three-dimensional tissues across diverse tumor types and patient studies. The collaborative project, which includes teams from the ETH Lausanne, the University Hospital of Zurich (USZ), and the University of Zurich (UZH), aims to strengthen precision oncology by improving how clinicians predict therapeutic approaches and disease progression. The findings were published on August 5, 2026, in the journal Nature.
Modern oncology relies heavily on understanding that a tumor is far more than a collection of rogue cells. Surrounding components—including immune cells, blood vessels, and other components of the surrounding tissue—actively shape how a cancer grows and responds to treatments. Even tumors with a similar cellular composition can react entirely differently to the exact same therapy. Pinpointing which cells are present, where they reside, and how they communicate with one another requires advanced techniques like spatial proteomics. Yet, the vast datasets generated by these methods have historically proved difficult to evaluate and compare across separate scientific studies.
Overcoming Data Fragmentation in Spatial Proteomics
To address these analytical roadblocks, the research team created the “Virtual Tissues” (VirTues) model. While most computational biology models are engineered to answer only a single, predetermined question, VirTues applies the logic of language-based foundation models to spatial biology. Rather than learning connections between words, the model learns relationships among proteins, cells, and their spatial contexts. This creates a unified representation applicable to a wide range of downstream analyses.
“Welche Zellen in einem Tumor vorhanden sind, ist nur ein Teil des Gesamtbildes,” noted Charlotte Bunne, who leads the research group at ETH Lausanne. “Wir müssen ebenso wissen, wo diese sie sich befinden und wie sie miteinander interagieren.”
Managing the sheer scale of tissue analysis requires robust computational frameworks. Andreas Wicki, deputy clinic director at the Clinic for Medical Oncology and Hematology at the USZ, emphasized the clinical necessity of these tools. “In der Onkologie ist die schiere Menge an Daten, die bei der Analyse von Gewebe anfallen, in den letzten Jahren exponentiell gewachsen,” Wicki stated, adding that computer-aided modeling provides the essential bridge for making complex data clinically useful for patients.
Key Advances and Massive Dataset Integration
VirTues achieves two primary technical milestones. First, it brings together the largest publicly available dataset of spatial proteomics measurements to date, encompassing more than 12,000 images from over 5,000 patients. Second, it implements a framework designed to learn and interpret intricate patterns across large datasets.
Johann Wenckstern, a doctoral student in Bunne’s group and the study’s lead author, highlighted the platform’s adaptability. “Diese Flexibilität ist eine der grössten Stärken von VirTues,” Wenckstern said. The framework allows laboratories to integrate datasets into a single model, analyze tissues from different studies in the same way, and swiftly interpret findings across diverse patient cohorts.
Because the system can process proteins measured across varied studies, cancer types, and assay panels, every new tissue sample fits into a standardized comparative framework. This lets incoming datasets build dynamically upon previously acquired learning.
Future Directions for Precision Oncology
For clinical researchers, the model serves as an analysis instrument. Bunne explained that scientists can leverage the tool to differentiate tumor patterns and identify biomarkers linked directly to disease progression and treatment responsiveness.
Looking ahead, investigators plan to transition these analytical models into a medical product. Wicki noted that integrating platforms like VirTues into local and national tumor boards for precision oncology represents a logical subsequent phase for modern precision cancer care, helping physicians evaluate multiple biomarkers simultaneously without restrictive assumptions about marker priority.
Keep reading