Nature paper unveils Virtual Tissues foundation model that resolves spatial proteomics across scales

August 5, 2026 at 05:00 AM1 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Researchers have introduced a general-purpose foundation model for spatial proteomics, Virtual Tissues (VirTues), in a study published in Nature. The model harmonises protein imaging data from heterogeneous platforms and cancer types, enabling robust biomarker discovery and patient stratification. The work addresses a long-standing challenge in the field, where custom panels and protocols have hindered knowledge transfer across studies.

The VirTues Foundation Model

Virtual Tissues (VirTues) is a marker-aware, multi-scale foundation model that learns representations of proteins, cells, niches and tissues directly from multiplex imaging data. It supports tasks such as marker reconstruction, cell segmentation and typing, niche annotation, and spatial biomarker discovery. Crucially, it achieves zero-shot annotation across heterogeneous antibody panels and datasets, a capability that conventional encoders lack because they assume a fixed marker set. The model was pretrained on a corpus of images from diverse cancer cohorts, though the exact number of training samples was not disclosed. In benchmarks, VirTues-derived biomarkers predicted response to anti-PD-L1 chemo-immunotherapy and stratified disease-free survival in triple-negative breast cancer, outperforming existing clinical stratification schemes.

Clinical Validation in Triple-Negative Breast Cancer

The team evaluated VirTues on data from triple-negative breast cancer patients. The model’s biomarkers predicted response to anti-PD-L1 chemo-immunotherapy in one cohort and stratified disease-free survival in an independent cohort. Compared to current clinical stratification and state-of-the-art biomarkers derived from the same datasets, VirTues demonstrated superior performance. These results suggest that spatial proteomics-based biomarkers, when harmonized by a foundation model, could improve patient selection for immunotherapy.