A unified framework for high-fidelity single-cell / near-cellular scale recovery in spatially-resolved multi-omics and microscopy
Suitable for pathology fundation models and high-fidelity downstream analysis,
valid in single-cell and near-cellular scale,
and nominate candidate genes while sharpening lesion boundaries and resolving microenvironmental changes
| Scope | Systems | Abbreviation |
|---|---|---|
| Mass spectrometry imaging (MSI) | Time-of-Flight Secondary Ion Mass Spectrometry | TOFSIMS (200 nm, 1 μm) |
| Mass spectrometry imaging (MSI) | Desorption Electrospray Ionization Mass Spectrometry | DESI-MSI (50 μm) |
| Mass spectrometry imaging (MSI) | Matrix-Assisted Laser Desorption/Ionization Mass Spectrometry | MALDI-MSI (50 μm) |
| Labeled mass spectrometry imaging (MSI) | Imaging Mass Cytometry | IMC (1 μm) |
| Spatially-resolved transcriptomics (SRT) | 10x Genomics Visium spatially-resolved transcriptomics | SRT (27.5 μm, 55 μm) |
| Spatially-resolved proteomics (SRP) | PhenoCycler-Fusion 2.0 spatially-resolved proteomics | SRP (1 μm) |
| Histological imaging | Hematoxylin and Eosin staining | H&E (20×, 40×) |
| Immunofluorescence imaging | Multiplex Immunofluorescence | mIF |
| Structured Illumination microscopy | Structured Illumination Microscopy | SIM |
| Immunohistochemical imaging | Immunohistochemistry | IHC |
| Pathology fundation model | Uni for cell type classifier | Uni |
| Histopathological Diffusion-based Stain Transfer | Using HistDiST predict IHC | HistDiST |
| H&E-based SRT Transfer | Using Path2Space predict SRT | Path2Space |
| Cross-scale bimodal correlation analysis | Association analysis and disease diagnostic | - |
| Multiscale multimodal co-localization analysis | SRT-MSI-cell type co-localization analysis | - |
| Real tissue samples near-cellular reconstruction | Kmeans for SRT cluster (k=9) | - |
Degradation of high-fidelity spatial information in biomedical imaging compromises analytical reliability, and a unified framework spanning microscopy and spatially-resolved multi-omics with robust generalizability and biologically faithful reconstruction, has not been established yet. Here, we present Uni-SOR, a unified framework. We validate Uni-SOR’s generalizability across multiple microscopy and spatially-resolved multi-omics systems with significant improvements across diverse restoration metrics. Remarkably, even with 93.75% high-frequency information loss, Uni-SOR still enables efficiently restoration and preserves concordance with over 90% area under the curve in cross-scale analysis. Together, we demonstrate that Uni-SOR consistently outperforms the existing methods in defocused imaging, super-resolution and sparse sampling reconstruction of microscopy and spatially-resolved multi-omics with heterogeneous profiles, and enables high-fidelity high-frequency information recovery to facilitate biological exploration.
Uni-SOR is designed not only for visual restoration, but also for downstream biological and computational analysis. The recovered outputs can support multiple downstream tasks.
- Cell segmentation with pathology foundation models
Recovered images can be used as inputs for pathology foundation models to improve cell segmentation in degraded tissue images. - IHC prediction from recovered H&E
Recovered H&E images can provide enhanced morphological information for IHC prediction and virtual staining tasks. - Joint analysis across MSI, SRT and H&E
Recovered MSI, SRT and H&E data can be integrated for multi-modal spatial analysis, enabling more reliable alignment between molecular signals and tissue morphology.
This repository currently provides lightweight demo files for SRP and SIM (Endoplasmic Reticulum, ER) tasks. For lightweight tasks, we provide free online access through our homepage.
After downloading the repository, run one of the demo scripts according to the task.
python "code/run sparse-sapmling demo.py"python "code/run super-resolution demo.py"python "code/run SIM demo.py"Before running a demo, modify the file paths and pretrained weight paths in the corresponding script.
INPUT_TIFF_PATH = "path/to/your/input"
MODEL_WEIGHTS = "path/to/pretrained/weights"Only SRP and SIM demo files are currently included in this repository.
If Uni-SOR is useful for your research, please cite our work.
@article{unisor2026,
title={Uni-SOR: unified framework for high-fidelity recovery in spatially-resolved multi-omics and microscop},
author={Hao Xu, Yu Zheng, Xiaopin Lai, Tianci Gao, Fancheng Tan, Zhongze Wang, Yiting Wu, Yucheng Dai, Fangmeng Fu, Liangkai Zheng, Guihua Wang, Song-Ling Wang, Mao Li, Tie Shen, Shu-Hai Lin},
journal={ },
year={2026}
}This project is released for academic research use. Please check the license file for details.
Uni-SOR (Unified Spatial Observation Reconstruction)
Unified spatial omics and microscopy recovery for high-fidelity biological discovery