GARDEN is a Python package for quantitative characterization and interpretation of rare spatial heterogeneity from spatial omics data.
Note: Before using the mclust algorithm, ensure that the mclust package is installed in R and that os.environ['R_HOME'] is configured with the correct path by following these steps:
conda create -n GARDEN_env python=3.8
conda activate GARDEN_env
conda install r-base
pip install rpy2==3.4.1
R --quiet --no-restore
install.packages('mclust')
Next, we will set up the environment required for GARDEN：
git clone [email protected]:Briskzxm/GARDEN.git
cd <Path_to_GARDEN>
python setup.py build
python setup.py install
pip install -r requirements.txt
Note: During the installation process, you might encounter issues with the installation of torch_sparse, torch_scatter, or torch_geometric. If this happens, you will need to manually download the .whl files from PyTorch Geometric WHL.
Once downloaded, install the files using the following command (replace <file_name>.whl with the actual filename of the .whl file):
pip install <file_name>.whl
Input data of GARDEN :
- The input files include various data formats, with
h5adbeing a representative example containing spatial transcriptomics data with spatial coordinates stored in.obsm[‘spatial’].
import scanpy as sc
from GARDEN import GARDEN
file_path = '/home/user/data/spatial_data.h5ad'
adata = sc.read(file_path)
The definition and training step of GARDEN are carried out as follows:
# model definition
model = GARDEN.GARDEN(adata,device=device)
# model training
adata = model.train()
Subsequently, clustering analysis can be performed using algorithms such as mclust or leiden.
# Apply clustering algorithm
from GARDEN.utils import clustering
clustering(adata, n_clusters)
Subsequently, clustering analysis can be performed using algorithms such as mclust or leiden.
For spatial transcriptome datasets with high memory demands, we used batch training to run the model.
# Batch Train
model = GARDEN.GARDEN_Batch(adata, datatype = 'HD')
adata = model.train_expand(batch_number = 10)
Please see the Jupyter notebook in the Tutorial folder. It includes several tutorials, providing examples across different species, sequencing technologies, and diseases.
This project is covered under the MIT License.