SpaCell has been developed for analysing spatial transcriptomics (ST) data, which include imaging data of tissue sections and RNA expression data across the tissue sections. The ST data add a novel spatial dimension to the traditional gene expression data from dissociated cells. SpaCell is desinged to integrates the two histopathological imaging and sequencing fields, with the ultimate aim to discover novel biology and to improve histopathological diagnosis in a quantitative and automated way.
SpaCell implements (deep) neural network (NN) models like a multi-input and multi-output autoencoder, transfer learning with or without fine tuning and residual and separable convolutional NN architectures to identify cell types or to predict disease stages. The NN integrates millions of pixel intensity values with thousands of gene expression measurements from spatially-barcoded spots in a tissue.
Prior to model training, SpaCell enables users for implement a comprehensive data preprocessing workflow to filter, combine, and normalise images and gene expression matrices.
SpaCell allows to map histopathological annotation to the original, high-resolution whole-slide-images. The mapped annotion is then used for evaluating model prediction results, thereby comparing model performance with real-life pathological diagnosis.
- Requirements:
[python 3.6+]
[TensorFlow 1.4.0]
[scikit-learn 0.18]
[keras 2.2.4]
[seaborn 0.9.0]
[opencv 4.1.1]
[pandas 0.25.0]
[pillow 6.1.0]
[python-spams 2.6.1]
[staintools 2.1.2]
[Others as specified in the requirements.yml file]
- Installation:
2.1 Build from sources
To meet the requirements, we recommend user to use either (1) conda environment:
# Download SapCell from GitHub and install all required packages:
git clone https://github.com/BiomedicalMachineLearning/Spacell.git
cd Spacell
conda env create -f requirements.yml
# To activate environment:
conda activate SpaCell
# To exit environment:
conda deactivate
2.2 Build Docker container:
# Download the SpaCell Docker image
docker pull biomedicalmachinelearning/spacell:latest
# Run Docker container
docker run \
-it \
-v /path/to/your/data:/home/Spacell/dataset/ \ # mount your local data directory to container
biomedicalmachinelearning/spacell:latest
2.3 Install from PyPi
pip install SpaCell
| Build Type | Status | Artifacts |
|---|---|---|
| Ubuntu Linux 18.04 | Conda, Docker, PyPI | |
| Centos Linux 6/7 | Conda, Docker, PyPI | |
| MacOS(Mojave/Catalina) | Conda, Docker, PyPI | |
| Windows 10 | Docker, PyPI |
* python-spams 2.6.1, a dependence of staintools 2.1.2, is currently not avaliable on Windows 10 platform. We recommend Windows 10 user to use Docker container instead of Conda environment.
config.py
- Specify the dataset directory and output directory.
- Specify model parameters.
python image_normalization.py
python count_matrix_normalization.py
python dataset_management.py
python spacell_classification.py
python spacell_clustering.py -i /path/to/one/image.jpg -l /path/to/iamge/tiles/ -c /path/to/count/matrix/ -e 100 -k 2 -o /path/to/output/
-eis number of training epochs-kis number of expected clusters
python spacell_validation.py -d /path/to/data -a annotation.png -w wsi.jpeg -m affine_tranformation_matrix.txt -o output_folder -k clustering_predictions.tsv -c annotation_colour_range
-cis annotation colour range thresholds - blue_low green_low red_low blue_upper green_upper red_low-tindicates that annotations are not closed paths, so spacell with try to close the paths-fdownscale factor if the input whole slide image has already been downscaled-sspot size, optional, usually set automatically
For evaluating the algorithm, ALS (Amyotrophic lateral sclerosis) dataset, prostate cancer dataset, and a high density spatial transcriptomic HDST dataset were used.
If you find Spacell useful in your research, please consider citing:
Xiao Tan, Andrew T Su, Minh Tran, Quan Nguyen (2019). SpaCell: integrating tissue morphology and spatial gene expression to predict disease cells. (Manuscript is currently under-review)
The software is under active development by the Biomedical Machine Learning Lab at the Institute for Molecular Bioscience (IMB, University of Queensland).
Please contact Dr Quan Nguyen (quan.nguyen@uq.edu.au), Andrew Su (a.su@uqconnect.edu.au), and Xiao Tan (xiao.tan@uqconnect.edu.au) for issues, suggestions, and we very welcome collaboration opportunities.