Spatial-MGCN: a novel multi-view graph convolutional network for identifying spatial domains with attention mechanism

Requirements

Python==3.8.13

numpy==1.20.0

pandas==1.4.4

scipy==1.8.1

stlearn==0.4.8

pytorch==1.11.0

torch_geometric==2.1.0

torch_sparse==0.6.15

torch_scatter==2.0.9

matplotlib==3.5.3

Example

1 Raw data

Raw data should be placed in the folder data.

we put the DLPFC dataset in data/DLPFC. Need to be extracted firstly.

For 10x Spatial Transcripts (ST) datasets, files should be put in the same structure with that provided by 10x website. Taking DLPFC for instance:

data/DLPFC/151507/

spatial/ # The folder where files for spatial information can be found

metadata.tsv # mainly for annotation

filtered_feature_bc_matrix.h5 # gene expression data

2 Configuration

The meaning of each argument in config.py is listed below.

--epochs: the number of training epochs.

--lr: the learning rate.

--weight_decay: the weight decay.

--k: the k-value of the k-nearest neighbor graph, which should be within {8...15}.

--radius：the spatial location radius.

--nhid1: the dimension of the first hidden layer.

--nhid2: the dimension of the second hidden layer.

--dropout: the dropout rate.

--alpha: the value of hyperparameter alpha for zinb_loss, which should be within {0.1,1,10}.

--beta: the value of hyperparameter beta for con_loss, which should be within {0.1,1,10}.

--gamma: the value of hyperparameter gamma for reg_loss, which should be within {0.1,1,10}.

--no_cuda: whether to use GPU.

--no_seed: whether to take the random parameter.

--seed: the random parameter.

--fdim: the number of highly variable genes selected.

3 Data Preprocessing and Graph Construction

Run Spatial-MGCN/DLPFC_generate_data.py to preprocess the raw DLPFC data:

python DLPFC_generate_data.py

Augments:

--savepath: the path to save the generated file.

For dealing other ST datasets, please modify the data name.

4 Usage

For training Spatial-MGCN model, run

'python DLPFC_test.py'

All results are saved in the result folder. We provide our results in the folder result for taking further analysis.

(1) The cell clustering labels are saved in Spatial_MGCN_idx.csv, where the first column refers to cell index, and the last column refers to cell cluster label.

(2) The trained embedding data are saved in Spatial_MGCN_emb.csv.

For Human_Breast_Cancer and Mouse_Olfactory datasets, the running process is the same as above. You just need to run the command:

'python HBC_test.py'

'python MOB_test.py'

Download all datasets used in Spatial-MGCN:

The datasets used in this paper can be downloaded from the following websites. Specifically,

(1) The LIBD human dorsolateral prefrontal cortex (DLPFC) dataset http://spatial.libd.org/spatialLIBD

(2) the processed Stereo-seq dataset from mouse olfactory bulb tissue https://github.com/JinmiaoChenLab/

(3) 10x Visium spatial transcriptomics dataset of human breast cancer https://support.10xgenomics.com/spatial-gene-expression/datasets/1.1.0/V1_Breast_Cancer_Block_A_Section_1

Additionally, the data employed in this study has been uploaded to Zenodo https://zenodo.org/records/10279295

Citation

Wang et al. Spatial-MGCN: a novel multi-view graph convolutional network for identifying spatial domains with attention mechanism. Briefings in Bioinformatics, 2023, 24(5): bbad262.