Volume 7,Issue 2
VirtualST: Morphology- and Cell-Composition-Conditioned Diffusion for Predicting Spatial Gene Expression from H&E Images
Spatial transcriptomics (ST) measures gene expression while preserving tissue spatial information, but its high cost and limited throughput restrict large-scale application. In contrast, hematoxylin and eosin (H&E)-stained histology images are widely available in routine pathology. We propose VirtualST, a conditional diffusion model for predicting spot-level spatial gene expression from H&E images. The model integrates histological features extracted by a pathology foundation model, local cell-type composition derived from nuclei segmentation, and spatial information from neighboring spots. VirtualST was evaluated on multiple cancer cohorts from HEST-bench and compared with representative histology-to-expression prediction methods. The results showed that VirtualST achieved competitive performance across different cancer types and performed well for representative colorectal cancer marker genes. These findings suggest that VirtualST provides an effective approach for spatial gene-expression prediction from routine histology images.
[1] Ståhl PL, Salmén F, Vickovic S, et al., 2016, Visualization and analysis of gene expression in tissue sections by spatial transcriptomics. Science, 353(6294): 78-82. DOI:10.1126/science.aaf2403.
[2] Jaume G, Doucet P, Song AH, et al., 2024, HEST-1k: A dataset for spatial transcriptomics and histology image analysis. Advances in Neural Information Processing Systems, 37. DOI:10.52202/079017-1704.
[3] He B, Bergenstråhle L, Stenbeck L, et al., 2020, Integrating spatial gene expression and breast tumour morphology via deep learning. Nature Biomedical Engineering, 4: 827-834. DOI:10.1038/s41551-020-0578-x.
[4] Pang M, Su K, Li M, 2021, Leveraging information in spatial transcriptomics to predict super-resolution gene expression from histology images in tumors. bioRxiv. DOI:10.1101/2021.11.28.470212.
[5] Zeng Y, Wei Z, Yu W, et al., 2022, Spatial transcriptomics prediction from histology jointly through Transformer and graph neural networks. Briefings in Bioinformatics, 23(5): bbac297. DOI:10.1093/bib/bbac297.
[6] Xie R, Pang K, Chung S, et al., 2023, Spatially resolved gene expression prediction from histology images via bi-modal contrastive learning. Advances in Neural Information Processing Systems, 36.
[7] Filiot A, Jacob P, Mac Kain A, et al., 2024, Phikon-v2, a large and public feature extractor for biomarker prediction. arXiv preprint, arXiv:2409.09173. DOI:10.48550/arXiv.2409.09173.
[8] Graham S, Vu QD, Raza SEA, et al., 2019, HoVer-Net: Simultaneous segmentation and classification of nuclei in multi-tissue histology images. Medical Image Analysis, 58: 101563. DOI:10.1016/j.media.2019.101563.
[9] Ho J, Jain AN, Abbeel P, 2020, Denoising diffusion probabilistic models. Advances in Neural Information Processing Systems, 33: 6840-6851.
[10] Liu J, Eckstein M, Wang Z, et al., 2026, Spatial transcriptomics expression prediction from histopathology based on cross-modal mask reconstruction and contrastive learning. Medical Image Analysis, 108: 103889. DOI:10.1016/j.media.2025.103889.
[11] van der Flier LG, Haegebarth A, Stange DE, et al., 2009, OLFM4 is a robust marker for stem cells in human intestine and marks a subset of colorectal cancer cells. Gastroenterology, 137(1): 15-17. DOI:10.1053/j.gastro.2009.05.035.
[12] Becker WR, Nevins SA, Chen DC, et al., 2022, Single-cell analyses define a continuum of cell state and composition changes in the malignant transformation of polyps to colorectal cancer. Nature Genetics, 54: 985-995. DOI:10.1038/s41588-022-01088-x.