STARS

Decoding Spatial Transcriptomics at Any Resolution: From Multicellular or Subcellular Spots to Individual Cells

We presented STARS (Spatial Transcriptomics across Any Resolution for Single Cells). Leveraging Vision Transformer model and contrastive learning, STARS combines high-resolution histology images with spot-level transcriptomics data to decode true single-cell gene expression from any multicellular or subcellular platforms. We demonstrated the advantage of our true single-cell method using public datasets and in-house datasets of mouse lung from 3 ST platforms (Visium, Visium HD and Stereo-seq). STARS was applied at tissue, individual cell, and molecular levels.

Framework

The code is licensed under the GPL-3.0 license.

1. Requirements

1.1 Operating systems:

The code in python has been tested on Linux.

1.2 Required packages in python:

anndata
numpy
opencv-python
pandas
python-louvain
rpy2
scanpy
scipy
seaborn
torch
torch-geometric
torchvision
tqdm
umap-learn

1.3 How to install STARS:
Before installing STARS, ensure that you have StarDist installed in your environment. If not, please follow the installation instructions here.

To download STARS, use the following command:

git clone https://github.com/Zhaocy-Research/STARS.git

(1) cd STARS

(2) conda create --name STARS python=3.9

(3) conda activate STARS  

STARS can be installed via pip using the following command:

pip install stars-omics

After installation, you can import the package in Python as:

import stars_omics

2. Instructions: Demo on mouse lung data.

We provide an example notebook, visium_06.ipynb, to implement the experimental results from the paper.

Data Access

https://pitt-my.sharepoint.com/:f:/g/personal/chz113_pitt_edu/EuGVB7q_xG1FtaTGj7PrW2wBNVADPt_9ZBJGOxnu0zdMwg?e=X1XMC1

Data can be accessed via the following link. Please download the data and update the data path in the code. The file name remains unchanged. For example, modify the code from:

img_fold = os.path.join('/ix1/wchen/Shiyue/Projects/2023_06_Influ_Mouse_Lung_ST/RawData/Fastq/Alcorn_Visium_FFPE_Images/', name + '.TIF')

to

img_fold = os.path.join('/your/local/path/', name, '.TIF')

Note: The path /ix1/wchen/Shiyue/Projects/2023_06_Influ_Mouse_Lung_ST/RawData/Fastq/Alcorn_Visium_FFPE_Images/ is just an example. Other data paths used in the code should also be modified to your local path in the same manner.

3. Instructions: Demo on nuclei segmentation.

We provide an example notebook, nuclei_segmentation.ipynb, to implement the experimental results from the paper.

Data Access

https://www.10xgenomics.com/products/visium-hd-spatial-gene-expression/dataset-human-crc

For nuclei segmentation, we use StarDist. Since StarDist requires TensorFlow to utilize the GPU version, it may cause conflicts with our current model. Therefore, we recommend that the user follow the original installation guidelines provided by the StarDist package before running our notebook (https://github.com/stardist/stardist).

4. Instructions: Demo on CRC data.

We provide an example notebook, CRC.ipynb, to implement the experimental results from the paper.

Data Access

https://www.10xgenomics.com/products/visium-hd-spatial-gene-expression/dataset-human-crc

Please download the data and update the data path in the code as shown in Instruction 2.