DeepSpaCE

The Deep learning model for Spatial gene Clusters and Expression (DeepSpaCE) is a method that predicts spatial gene-expression levels and transcriptomic cluster types from tissue section images using deep learning.

Note

DeepSpaCE is now beta version. Please note that we shall not be responsible for any loss, damages and troubles.

Requirement
Installation
Usage
Citation
License
FAQ
Release notes

Requirement

Singularity (verified in v3.7)

Installation

Clone the DeepSpaCE repository

git clone https://github.com/tmonjo/DeepSpaCE

Build a Singularity image

Build an image on your local environment since root privileges are required. Then, you can run DeepSpaCE with "DeepSpaCE.sif" on any servers.

sudo singularity build DeepSpaCE.sif DeepSpaCE.srecipe

Usage

Input files (all files must be located in a same directory of sampleName)

Space Ranger outputs

/home/$USER/DeepSpaCE/data/{sampleName}/SpaceRanger/analysis/

/home/$USER/DeepSpaCE/data/{sampleName}/SpaceRanger/spatial/

/home/$USER/DeepSpaCE/data/{sampleName}/SpaceRanger/filtered_feature_bc_matrix.h5
TIFF image (same directory)

/home/$USER/DeepSpaCE/data/{sampleName}/SpaceRanger/image.tif

Preprocessing 1: Section image files

singularity exec DeepSpaCE.sif \
    python script/CropImage.py \
        --dataDir /home/$USER/DeepSpaCE/data \
        --sampleName Human_Breast_Cancer_Block_A_Section_1 \
        --transposeType 0 \
        --radiusPixel 75 \
        --extraSize 150 \
        --quantileRGB 80

Preprocessing 2: Satial expression data measured by Visium

singularity exec DeepSpaCE.sif \
    Rscript script/NormalizeUMI.R \
        --dataDir /home/$USER/DeepSpaCE/data \
        --sampleName Human_Breast_Cancer_Block_A_Section_1 \
        --threshold_count 1000 \
        --threshold_gene 1000

Run DeepSpaCE (Training and validation) (running on GPU is recommended)

singularity exec --nv DeepSpaCE.sif \
    python script/DeepSpaCE.py \
        --dataDir /home/$USER/DeepSpaCE/data \
        --outDir /home/$USER/DeepSpaCE/out \
        --sampleNames_train Human_Breast_Cancer_Block_A_Section_1 \
        --sampleNames_test Human_Breast_Cancer_Block_A_Section_1 \
        --sampleNames_semi None \
        --semi_option normal \
        --seed 0 \
        --threads 8 \
        --GPUs 1 \
        --cuda \
        --transfer \
        --model VGG16 \
        --batch_size 128 \
        --num_epochs 10 \
        --lr 1e-4 \
        --weight_decay 1e-4 \
        --clusteringMethod graphclust \
        --extraSize 150 \
        --quantileRGB 80 \
        --augmentation flip,crop,color,random \
        --early_stop_max 5 \
        --cross_index 0 \
        --geneSymbols ESR1,ERBB2,MKI67

Super-resolution

Run super-resolution

singularity exec DeepSpaCE.sif \
    python script/SuperResolution.py \
        --dataDir /home/$USER/DeepSpaCE/data \
        --outDir /home/$USER/DeepSpaCE/out \
        --sampleName Human_Breast_Cancer_Block_A_Section_1 \
        --model VGG16 \
        --seed 0 \
        --threads 8 \
        --modelName teacher \
        --batch_size 128 \
        --extraSize 150 \
        --quantileRGB 80 \
        --geneSymbols ESR1,ERBB2,MKI67

Plot a super-resolved image

singularity exec DeepSpaCE.sif \
    Rscript script/PlotSuperResolution.R \
        --dataDir /home/$USER/DeepSpaCE/data \
        --outDir /home/$USER/DeepSpaCE/out \
        --sampleName Human_Breast_Cancer_Block_A_Section_1 \
        --geneSymbol ESR1 \
        --extraSize 150

Semi-supervised learning (under development)

Prepare image_list.txt for semi-supervised learning. image_list.txt should contain "ImageSet", "sample_No, and "image_path".

/home/$USER/DeepSpaCE/data/Visium/ImageSet/"+sampleName+"/ImageSet_0/image_list.txt

Citation

Monjo, T., Koido, M., Nagasawa, S. et al. Efficient prediction of a spatial transcriptomics profile better characterizes breast cancer tissue sections without costly experimentation. Sci Rep 12, 4133 (2022). https://doi.org/10.1038/s41598-022-07685-4

License

GNU General Public License v3.0

FAQ

Can I install DeepSpaCE without Singularity?

Please install Python 3.6, R >= 4.1, and libraries written in "DeepSpaCE.srecipe".

Pipfile is also available. (Pipenv)
```
 pipenv install Pipfile
```

Release notes

v0.1 (November 14 2021): First release

baranwa2 / DeepSpaCE