MaskGraphene: Advancing joint embedding, clustering, and batch correction for spatial transcriptomics using graph-based self-supervised learning

Implementation for Recomb-seq 2024 paper: MaskGraphene.

Online documentation

Please refer to this page. We hope this could help to better explore and investigate this tool.

Dependencies

• Python >= 3.9
• Pytorch == 2.0.1
• anndata==0.9.2
• h5py==3.9.0
• hnswlib==0.7.0
• igraph==0.10.8
• matplotlib==3.6.3
• paste-bio==1.4.0
• POT==0.9.1
• rpy2==3.5.14
• scanpy==1.9.1
• umap-learn==0.5.4
• wandb
• pyyaml == 5.4.1

Installation

conda create -n MaskGraphene python=3.9 

conda activate MG

git clone https://github.com/OliiverHu/MaskGraphene.git

pip install -r requirements.txt

For DGL package, please refer to link

pip install  dgl -f https://data.dgl.ai/wheels/cu117/repo.html
pip install  dglgo -f https://data.dgl.ai/wheels-test/repo.html

Quick Start

For quick start, you could run the scripts:

mouse Hypothalamus -0.19/-0.24 generate hard-links

python ../localMG_main.py --max_epoch 3000 --max_epoch_triplet 1000 --logging False --section_ids " -0.19,-0.24" --num_class 8 --load_model False --num_hidden "512,32" 
                          --exp_fig_dir "./" --h5ad_save_dir "./" --st_data_dir "./" --alpha_l 3 --lam 1 --loss_fn "sce" --mask_rate 0.50 --in_drop 0 --attn_drop 0 --remask_rate 0.50
                          --seeds 2023 --num_remasking 1 --hvgs 0 --dataset mHypothalamus --consecutive_prior 1 --lr 0.001

mouse Hypothalamus -0.19/-0.24

python ../maskgraphene_main.py --max_epoch 3000 --max_epoch_triplet 1000 --logging False --section_ids " -0.19,-0.24" --num_class 8 --load_model False --num_hidden "512,32" 
                               --exp_fig_dir "./" --h5ad_save_dir "./" --st_data_dir "./" --alpha_l 3 --lam 1 --loss_fn "sce" --mask_rate 0.50 --in_drop 0 --attn_drop 0 --remask_rate 0.50
                               --seeds 2023 --num_remasking 1 --hvgs 0 --dataset mHypothalamus --consecutive_prior 1 --lr 0.001

DLPFC 151507/151508 generate hard-links

python ../localMG_main.py --max_epoch 2000 --max_epoch_triplet 500 --logging False --section_ids "151507,151508" --num_class 7 --load_model False --num_hidden "512,32"
                          --exp_fig_dir "./" --h5ad_save_dir "./" --st_data_dir "./"  --alpha_l 1 --lam 1 --loss_fn "sce" --mask_rate 0.5 --in_drop 0 --attn_drop 0 --remask_rate 0.1
                          --seeds 2023 --num_remasking 1 --hvgs 3000 --dataset DLPFC --consecutive_prior 1 --lr 0.001

DLPFC 151507/151508

python ../maskgraphene_main.py --max_epoch 2000 --max_epoch_triplet 500 --logging False --section_ids "151507,151508" --num_class 7 --load_model False --num_hidden "512,32" 
                               --exp_fig_dir "./" --h5ad_save_dir "./" --st_data_dir "./" --alpha_l 1 --lam 1 --loss_fn "sce" --mask_rate 0.5 --in_drop 0 --attn_drop 0 --remask_rate 0.1 
                               --seeds 2023 --num_remasking 1 --hvgs 3000 --dataset DLPFC --consecutive_prior 1 --lr 0.001

Supported ST datasets:

• 10x Visium: DLPFC, Mouse Sagittal Brain
• Others: mouse Hypothalamus, Embryo

Notebook Tutorials

Tutorial 1: hard-links generation

Tutorial 2: MG on DLPFC

Tutorial 3: MG on MHypo

Tutorial 4: reproduction

Citing

Currently under review