The Virtual Tissues foundation model resolves spatial proteomics across scales.

VirTues, a foundation model for spatial proteomics, accurately predicts therapy response and patient outcomes in triple-negative breast cancer with 2–3 times higher accuracy than existing methods.

• Why it matters: Understanding tissue architecture at the molecular level is crucial for improving cancer diagnosis and treatment, but current methods lack robustness and transferability across different datasets and protocols.
• What they did: The team developed VirTues, a versatile model trained on multiplex imaging data that learns multi-scale, marker-aware representations, enabling tasks like marker reconstruction, cell typing, and biomarker discovery across diverse datasets.
• The result: VirTues-derived biomarkers outperform existing markers and clinical schemes in predicting immunotherapy response and survival, paving the way for more reliable, scalable spatial proteomics analyses.