Identifying batch-integrated domains from spatial transcriptomics via graph autoencoder with contrastive learning based on cross-modality and data augmentation

The structure of datasets and the weighted models To illustrate the dataset organization, we use DLPFC-151674 as an example. Under the folder "./Dataset/DLPFC/", the sample 151674 is structured as follows: ./Dataset/DLPFC/151674/│── spatial/ # spatial metadata and histology images│ ├── scalefactors_json.json│ ├── tissue_full_image.tif│ ├── tissue_hires_image.png│ ├── tissue_lowres_image.png│ ├── tissue_positions_list.csv│ ├── tissue_positions_list.txt│ │── filtered_feature_bc_matrix.h5 # filtered gene expression matrix│── 151674.pth # PyTorch weighted feature file│── 151674.npy # NumPy feature file│── truth.txt # ground truth labels Moreover, we have placed all the involved weight files for spatial transcriptomics datasets under the specified directory path for convenient access and reproducibility. ./Dataset/DLPFC/151674/151674.pth ./Dataset/DLPFC/subject/subject3.pth ./Dataset/v10x/cross-region/AP.pth ./Dataset/Mouse/cross_platform/CrossPlatform.pth ./Dataset/Mouse/Mouse_EMbryo/MouseEmbryo.pth Since the spatial coordinates across cross-dataset single-cell transcriptomics data are not aligned, direct visualization may lead to misleading presentation of the results. Therefore, we provide the transcriptomics data with aligned spatial coordinates as follows. ./Dataset/v10x/cross-region/adata_crossregion.h5ad ./Dataset/Mouse/cross_platform/adata_crossplafform.h5ad ./Dataset/Mouse/Mouse_EMbryo/adata_EmbyroTime.h5ad