A method to segment Spatial Transcriptomics data using scribble annotated histology image, or using output of other possibly non-spatial segmentation method (e.g. mclust).

Results are generated using Google Colab Standard GPU. To ensure reproducibility, the following is done:

torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False

Note that, you do not need to set these if you run the program in cpu instead of gpu.

Recommended Python version: 3.10.6

First set and activate your environment by using the following command:

conda env create -f environment.yml
conda activate scribble_seg

or, install the required packages by using the following command:

pip install -r requirements.txt

To run ScribbleSeg, you need 3 input files for a sample:

• The .h5 file containing the Anndata object
  • It must contain the array_row, and array_col representing the position of each spot in a grid
• A CSV file containing the principal component values
• A CSV file containing the scribble information for each spot

Place your .h5 file in ./Data/[dataset name]/[sample name]/reading_h5/. The current implementation reads the Anndata object by using scanpy.read_visium function. Hence, it expects a folder named spatial, containing the scalefactors, tissue high resolution image, low resolution image, tissue possitions list and so on, inside the reading_h5 folder. Depending on the structure of the .h5 file, you can use scanpy.read_h5ad fucntion to read the .h5 file, which doesn't require this spatial folder.

Place the csv containing the principal compoents in ./Data/[dataset_name]/[sample_name]/Principal_Components/CSV/. Make sure to change the name of the pc_csv_path variable according to the name of your CSV file in preprocessor.py.

Place the csv containing the principal compoents in ./Data/[dataset_name]/[sample_name]/. Make sure to change the name of the scr_csv_path variable according to the name of your CSV file in preprocessor.py.

Set the input parameters (e.g. hyperparameters, output directory name, and so on...) in ./Inputs/[file_name].json by following the example json files shown inside ./Inputs.

Scribbles can be generated using Loupe browser

After setting up the input parameters on ./Inputs/[file_name].json, the following steps are required to run ScribbleSeg:

1. At first you have to run preprocessor.py. For expert scribble scheme, run the following:

python preprocessor.py --scheme expert

Or, for mclust scribble scheme, run the following:

python preprocessor.py --scheme mclust

2. Then, to generate the segmentations for expert scribble scheme, run:

python expert_scribble_pipeline.py --params ./Inputs/expert_scribble_scheme_input.json

Or, to generate the segmentations for mclust scribble scheme, run:

python mclust_scribble_pipeline.py --params ./Inputs/mclust_scribble_scheme_input.json

3. Results will be put in Outputs directory.
4. To calculate adjusted rand index (ARI), you will need the ground truth labels. Put the ground truth labels at ./Data/[dataset name]/[sample name]/manual_annotations.csv.

The folder Supplementary_figures has the high quality figures of the supplementary information of our research paper.