A. Overview

SuperST is a novel algorithm that employs a deep image prior (i.e., Unet) to create dense matrices from low-resolution Visium Spatial Transcriptomics (ST) libraries. The details of how SuperST is run are explained in this repository and are also discussed in our paper. Briefly, SuperST is composed of five units, that said A, B, C, D, and E, which indicate an input H&E image, conceptual down-sampling unit, ‘concatenate’, conceptual up-sampling unit, and and the output of U-Net, respectively.

Looking forward, SuperST holds the potential to be utilized across various research domains that employ ST in imaging analysis workflows. Also, SuperST-produced image data could be effortlessly integrated with other spatial omics technologies. Overall, SuperST marks a significant stride forward in the realm of Spatial Transcriptomics, paving the way for a deeper understanding of intricate biological systems.

Publication

• Park, J.B. et al. (2023). Generation of Super-resolution Images from Barcode-based Spatial Transcriptomics Using Deep Image Prior. bioRxiv. [Link]

Patent

• Patent and Trademark Office, Republic of Korea. Application Number: 10-2023-0060674. Application Date: 2023.05.10. Super resolution Spatial transcriptome image generating method, Apparatus performing the same, and Program recorded in computer readable media for performing the same.

Poster

• Park, J., Cook, S., Lee, D., Choi, J., Yoo, S., Im, H.-J., Lee, D., & Choi, H. (2023). Cancer region definition using spatial gene expression patterns by super-resolution reconstruction algorithm for spatial transcriptomics data. Paper presented at the AACR Annual Meeting 2024, San Diego Convention Center, California, USA.

---

B. Environments

Download the superst.yml file and run the following command in an Anaconda Prompt:

conda env create -f superst.yml
conda activate SuperST

---

C. Data

When specifying tissue_dir, ensure the directory includes five crucial files typically produced by the 10X Visium platform:

• filtered_feature_bc_matrix.h5
• spatial/tissue_positions_list.csv
• spatial/scalefactors_json.json
• spatial/tissue_lowres_image.png
• spatial/tissue_hires_image.png

---

D. How to Use 1

step 1. Open a Linux terminal.

step 2. Git clone SuperST at the Linux terminal.

git clone https://github.com/portrai-io/SuperST.git

step 3. Make a jupyter notebook file at the same directory as app.py.

step 4. Open your jupyter notebook or jupyter lab.

step 5. Import all the functions from app.py into a Jupyter notebook file, and use the functions with the following manner.

from app import *
tissue_dir = './Data/10X/V1_Adult_Mouse_Brain_Coronal_Section_2/'
Tgenes = ['Rbfox3', 'Lamp5']
demask_image_t, demask_image_t_g, tsimg1_, conv_spatial_images_ = \\
    merge(tissue_dir, Tgenes, num_iter = 32, is_cut_bg = True)

Here, demask_image_t indicates SuperST images assuming zero-diffusion, while demask_image_t_g means SuperST images with a modifiable Gaussian kernel size specified by kernel_gauss.

---

E. How to Use 2

step 1. Open a Linux terminal.

step 2. Git clone SuperST at the Linux terminal.

git clone https://github.com/portrai-io/SuperST.git

step 3. Type a command like below.

python app.py \\
--tissue_dir ./Data/10X/V1_Adult_Mouse_Brain_Coronal_Section_2/ \\
--Tgenes Rbfox3 Lamp5 \\
--num_iter 32 \\
--is_cut_bg True

step 4. The outfile, SuperST.hdf5, at the same directory as app.py can be open with the following in a jupyter notebook file.

f = h5py.File("SuperST.hdf5", 'r')
dset1 = f['demask_image_t'] 
dset2 = f['demask_image_t_g'] 
dset3 = f['tsimg1_'] 
dset4 = f['conv_spatial_images_']

Here, demask_image_t indicates SuperST images assuming zero-diffusion, while demask_image_t_g means SuperST images with a modifiable Gaussian kernel size specified by kernel_gauss.

---

F. Example

Refer to the jupyter notebook file at the Example directory, where the usage methods are briefly introduced. The example dataset originates from the 10x Genomics Dataset (ref).

---

G. Contact

We, as Portrai. Inc., innovate the process of developing new drugs beyond the limits of human cognition, and deliver safer and more effective new drugs to humanity. For any questions or inquiries, please contact us at contact@portrai.io.