Spatial transcriptomics (ST) enables gene expression mapping within anatomical context but remains costly and low-throughput. Hematoxylin and eosin (H\&E) staining offers rich morphology yet lacks molecular resolution. We present \textbf{\ours} (\textbf{S}patial \textbf{T}ranscriptomics and hist\textbf{O}logy \textbf{R}epresentation \textbf{M}odel), a foundation model trained on 1.2 million spatially resolved transcriptomic profiles with matched histology across 18 organs. Using a hierarchical architecture integrating morphological features, gene expression, and spatial context, STORM bridges imaging and omics through robust molecular--morphological representations. STORM enhances spatial domain discovery, producing biologically coherent tissue maps, and outperforms existing methods in predicting spatial gene expression from H\&E images across 11 tumor types. The model is platform-agnostic, performing consistently across Visium, Xenium, Visium HD, and CosMx. Applied to 23 independent cohorts comprising 7,245 patients, STORM significantly improves immunotherapy response prediction and prognostication over established biomarkers, providing a scalable framework for spatially informed discovery and clinical precision medicine.
@article{arxiv.2604.03630,
title = {A Multimodal Foundation Model of Spatial Transcriptomics and Histology for Biological Discovery and Clinical Prediction},
author = {Jinxi Xiang and Siyu Hou and Yuchen Li and Ryan Quinton and Xiaoming Zhang and Feyisope Eweje and Xiangde Luo and Yijiang Chen and Zhe Li and Colin Bergstrom and Ted Kim and Sierra Willens and Francesca Maria Olguin and Matthew Abikenari and Andrew Heider and Sanjeeth Rajaram and Joel Neal and Maximilian Diehn and Xiang Zhou and Ruijiang Li},
journal= {arXiv preprint arXiv:2604.03630},
year = {2026}
}
Comments
29 pages, 5 figures. This manuscript is a work in progress; further updates and revisions will be posted as they become available