Spatially resolved sequencing technologies have revolutionized the characterization of biological regulatory processes within microenvironment by simultaneously accessing the states of genomic regions, genes and proteins, along with the spatial coordinates of cells, necessitating advanced computational methods for the cross-modality and multi-sample integrated analysis of spatial omics datasets. To address this gap, we propose PRESENT, an effective and scalable contrastive learning framework, for the cross-modality representation of spatially resolved omics data. Through comprehensive experiments on massive spatially resolved datasets, PRESENT achieves superior performance across various species, tissues, and sequencing technologies, including spatial epigenomics, transcriptomics, and multi-omics. Specifically, PRESENT empowers the incorporation of spatial dependency and complementary omics information simultaneously, facilitating the detection of spatial domains and uncovering biological regulatory mechanisms within microenvironment. Furthermore, PRESENT can be extended to the integrative analysis of horizontal and vertical samples across different dissected regions or developmental stages, thereby promoting the identification of hierarchical structures from a spatiotemporal perspective.
numpy==1.24.4
pandas==2.0.3
scipy==1.9.3
scikit-learn==1.3.2
louvain==0.8.0,
leidenalg==0.9.1,
anndata==0.9.2
networkx==3.1
scanpy==1.9.8
episcanpy==0.3.2
genomicranges==0.4.2
iranges==0.2.1
biocutils==0.1.3
torch==2.0.0
torch-geometric==2.3.1
PRESENT is available on PyPI here and can be installed via
pip install bio-present
pip install pyg_lib==0.2.0 torch_sparse==0.6.17 torch_cluster==1.6.1 torch_scatter==2.1.1 -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
You can also install PRESENT from GitHub via
git clone https://github.com/lizhen18THU/PRESENT.git
cd PRESENT
python setup.py install
pip install pyg_lib==0.2.0 torch_sparse==0.6.17 torch_cluster==1.6.1 torch_scatter==2.1.1 -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
The input of different spatial omics layers should be raw count matrices in anndata.AnnData format, where spatial coordinate matrix is stored in the adata.obsm[spatial_key] of corresponding AnnData object. The output including the joint representations and the identified domains, stored in an AnnData file and two csv files.
Suppose we have a spatial transcriptomics data, where the RNA raw count matrix and the spatial coordinate matrix are stored in file data/adata_rna.h5ad. The command to run PRESENT model on the spatial multi-omics data is
python3 PRESENT_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_rna_path data/adata_rna.h5ad --gene_min_cells 1 --num_hvg 3000 --nclusters 10
Optionally, PRESENT also can integrate reference data to address datasets with low sequencing depth and signal-to-noise ratio. Suppose the reference transcriptomic data are stored in data/rdata_rna.h5ad and the annotation is stored in the rdata_rna.obs["domains"], then the command to run PRESENT model is
python3 PRESENT_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_rna_path data/adata_rna.h5ad --rdata_rna_path data/rdata_rna.h5ad --rdata_rna_anno domains --gene_min_cells 1 --num_hvg 3000 --nclusters 10
Similarly, the command to run PRESENT model on the spatial ATAC-seq data is
python3 PRESENT_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_atac_path data/adata_atac.h5ad --peak_min_cells_fraction 0.03 --nclusters 10
Optionally,
python3 PRESENT_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_atac_path data/adata_atac.h5ad --rdata_rna_path data/rdata_rna.h5ad --rdata_rna_anno domains --peak_min_cells_fraction 0.03 --nclusters 10
Suppose we have a spatial RNA-ATAC co-profiling data, where the RNA and ATAC raw count matrix are stored in paired files data/adata_rna.h5ad and data/adata_atac.h5ad.
python3 PRESENT_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_rna_path data/adata_rna.h5ad --gene_min_cells 1 --num_hvg 3000 --adata_atac_path data/adata_atac.h5ad --peak_min_cells_fraction 0.03 --nclusters 10
Similarly, for spatial RNA-ADT co-profiling data, the command denotes
python3 PRESENT_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_rna_path data/adata_rna.h5ad --gene_min_cells 1 --num_hvg 3000 --adata_adt_path data/adata_adt.h5ad --protein_min_cells 1 --nclusters 10
Suppose we have two spatial transcriptomics samples for integrative analysis, where the two RNA raw count matrices are stored in separated AnnData files data/adata_rna1.h5ad and data/adata_rna2.h5ad. The command to run PRESENT-BC model is
python3 PRESENT_BC_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_rna_path_list data/adata_rna1.h5ad data/adata_rna2.h5ad --gene_min_cells 1 --num_hvg 3000 --nclusters 10
If the feature sets of different samples have already been unified and the expression matrices have been concatnated into a single AnnData file data/adata_rna_concatnated.h5ad, and the batch indices are stored in 'adata_rna_concatnated.obs["batch"]', then
python3 PRESENT_BC_script.py --outputdir ./PRESENT_output --spatial_key spatial --batch_key batch --adata_rna_path_list data/adata_rna_concatnated.h5ad --gene_min_cells 1 --num_hvg 3000 --nclusters 10
Similarly, the command to run PRESENT-BC model for the integration of spatial ATAC-seq data is
python3 PRESENT_BC_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_atac_path_list data/adata_atac1.h5ad data/adata_atac2.h5ad --peak_min_cells_fraction 0.03 --nclusters 10
and
python3 PRESENT_BC_script.py --outputdir ./PRESENT_output --spatial_key spatial --batch_key batch --adata_atac_path_list data/adata_atac_concatnated.h5ad --peak_min_cells_fraction 0.03 --nclusters 10
Suppose we have a spatial RNA-ATAC co-profiling data, where the RNA and ATAC raw count matrix of different samples are stored in files data/adata_rna1.h5ad, data/adata_rna2.h5ad, data/adata_atac1.h5ad and data/adata_atac2.h5ad. Then
python3 PRESENT_BC_script.py --outputdir ./PRESENT_output --spatial_key spatial --adata_rna_path_list data/adata_rna1.h5ad data/adata_rna2.h5ad --gene_min_cells 1 --num_hvg 3000 --adata_atac_path_list data/adata_atac1.h5ad data/adata_atac2.h5ad --peak_min_cells_fraction 0.03 --nclusters 10
If the feature sets of different samples have already been unified and the expression matrices of each omics layer have been concatnated into a single AnnData file, then
python3 PRESENT_BC_script.py --outputdir ./PRESENT_output --spatial_key spatial --batch_key batch --adata_rna_path_list data/adata_rna_concatnated.h5ad --gene_min_cells 1 --num_hvg 3000 --adata_atac_path_list data/adata_atac_concatnated.h5ad --peak_min_cells_fraction 0.03 --nclusters 10
- {--outputdir}: A path specifying where the final results are stored, default: ./PRESENT_output
- {--spatial_key}: adata.obsm key under which to load the spatial matrix in the AnnData object, default: spatial
- {--batch_key}: adata.obs key under which to load the batch indices in the AnnData object, default: batch
- {--nclusters}: Number of spatial clusters, default: 10
- {--gene_min_cells}: Minimum number of cells expressed required for a gene to pass filtering, default: 1
- {--peak_min_cells_fraction}: Minimum fraction of cells accessible required for a peak to pass filtering, default: 0.03
- {--protein_min_cells}: Minimum number of cells expressed required for a protein to pass filtering, default: 1
- {--num_hvg}: Number of highly variable genes to select for RNA data, default: 3000
- {--d_lat}: The latent dimension of final embeddings, default: 50
- {--k_neighbors}: Number of neighbors for each spot to construct graph, default: 6
- {--epochs}: Max epochs to train the model, default: 100
- {--lr}: Initial learning rate, default: 0.001
- {--batch_size}: Batch size for training, default: 320
- {--device}: Device used for training (cuda or cpu), default: cuda
- {--device_id}: Which gpu to use for training
Zhen Li, Xuejian Cui, Xiaoyang Chen, Zijing Gao, Yuyao Liu, Yan Pan, Shengquan Chen and Rui Jiang. "Cross-modality representation and multi-sample integration of spatially resolved omics data." Preprint at bioRxiv https://doi.org/10.1101/2024.06.10.598155 (2024).
