Inferring Supper-Resolution Tissue Architecture by Integrating Spatial Transcriptomics and Histology
This software package implements iStar (Inferring Supper-resolution Tissue ARchitecture), which enhances the spatial resolution of spatial transcriptomic data from a spot-level to a near-single-cell level. The iStar method is presented in the following paper:
Daiwei Zhang, Amelia Schroeder, Hanying Yan, Haochen Yang, Jian Hu, Michelle Y. Y. Lee, Kyung S. Cho, Katalin Susztak, George X. Xu, Michael D. Feldman, Edward B. Lee, Emma E. Furth, Linghua Wang, Mingyao Li. Inferring super-resolution tissue architecture by integrating spatial transcriptomics with histology. Nature Biotechnology (2024). https://doi.org/10.1038/s41587-023-02019-9
To run the demo,
# Use Python 3.9 or above
pip install -r requirements.txt
./run_demo.sh
Using GPUs is highly recommended.
he-raw.jpg
: Raw histology imagecnts.tsv
: Gene count matrix.- Row 1: Gene names.
- Row 2 and after: Each row is a spot.
- Column 1: Spot ID.
- Column 2 and after: Each column is a gene.
locs-raw.tsv
: Spot locations- Row 1: Header
- Row 2 and after: Each row is a spot. Must match rows in
cnts.tsv
- Column 1: Spot ID
- Column 2: x-coordinate (horizontal axis). Must be in the same space as axis-1 (column) of the array indices of pixels in
he-raw.jpg
. - Column 2: y-coordinate (vertical axis). Must be in the same space as axis-0 (row) of the array indices of pixels in
he-raw.jpg
.
pixel-size-raw.txt
: Side length (in micrometers) of pixels inhe-raw.jpg
. This value is usually between 0.1 and 1.0.- For Visium data, this value can be approximated by
8000 / 2000 * tissue_hires_scalef
, wheretissue_hires_scalef
is stored inscalefactors_json.json
.
- For Visium data, this value can be approximated by
radius-raw.txt
: Number of pixels per spot radius inhe-raw.jpg
.- For Visium data, this value can be computed by
spot_diameter_fullres * 0.5
, wherespot_diameter_fullres
is stored inscalefactors_json.json
, and should be close to55 * 0.5 / pixel_size_raw
.
- For Visium data, this value can be computed by
The software package is licensed under GPL-3.0. For commercial use, please contact Daiwei (David) Zhang and Mingyao Li.
The codes for iStar are written by Daiwei (David) Zhang and under active development. Please open an issue on GitHub if you have any questions about the software package.
The codebase for the hierarchical vision transformer is built upon Vision Transformer (as implemented by Hugging Face), DINO, and HIPT. We thank the authors for releasing the codes and the model weights.
If you find this work useful, please consider citing
@article{zhang2024inferring,
title = {Inferring Super-Resolution Tissue Architecture by Integrating Spatial Transcriptomics with Histology},
author = {Zhang, Daiwei and Schroeder, Amelia and Yan, Hanying and Yang, Haochen and Hu, Jian and Lee, Michelle Y. Y. and Cho, Kyung S. and Susztak, Katalin and Xu, George X. and Feldman, Michael D. and Lee, Edward B. and Furth, Emma E. and Wang, Linghua and Li, Mingyao},
year = {2024},
month = jan,
journal = {Nature Biotechnology},
pages = {1--6},
doi = {10.1038/s41587-023-02019-9},
}
as well as Vision Transformer, DINO, and HIPT.