SpatialPrompt : Spatially aware scalable and accurate tool for spot deconvolution and clustering in spatial transcriptomics

Article url: https://doi.org/10.1038/s42003-024-06349-5
Cite the article :

Swain, A.K., Pandit, V., Sharma, J. et al. SpatialPrompt: spatially aware scalable and accurate tool for spot deconvolution and domain identification in spatial transcriptomics. Commun Biol 7, 639 (2024). https://doi.org/10.1038/s42003-024-06349-5 

SpatialPromptDB database url: https://swainasish.github.io/SpatialPrompt/

Graphical overview

SpatialPrompt integrates gene expression, spatial location, and reference single cell RNA sequencing data for spatial deconvolution.
For spatial clustering, tool requires only gene expression with spatial locations.

Installation using PIP

Platforms: Ubuntu: 22.04/20.04, Windows: 10/11, Mac-OS: Ventura - Python: 3.10 Preferable

pip install spatialprompt

Tutorials and Reproducibility section

Tutorial-1 (10X Visium human DLPFC dataset) (Figure 2)

Tutorial-2 (10X Visium mouse cortex dataset) (Figure 3)

Quick start

Spatial spot deconvolution

For spot deconvolutiom SpatialPrompt requires:

• sc_array: Matrix of Single-cell data, where rows are the cells and columns are the genes.
• st_array: Matrix of Spatial data, where rows are the cells and columns are the genes.
• sc_genes: Gene names of the sc_array matrix.
• st_genes: Gene names of the st_array matrix.
• sc_labels: Cell type annotations of sc_array.
• x_cord: X coordinate array of spatial data.
• y_cord: Y coordinate array of spatial data.

Import the library and download the data

import scanpy as sc
import spatialprompt as sp
import gdown
gdown.download("https://drive.google.com/uc?id=1n3ACaWjfjXJ8P6IJhxSUax_vBagIUPLV","sc_m_cortex.h5ad", quiet=False)
gdown.download("https://drive.google.com/uc?id=1h7E5nPs2ga1ixOBDjLJKK7V8xJq93ez5","st_m_cortex.h5ad", quiet=False)

Load the data into Scanpy and retrieve required information

sc_data = sc.read_h5ad("sc_m_cortex.h5ad")
st_data = sc.read_h5ad("st_m_cortex.h5ad")
#Requires raw data
sc_array = sc_data.X
sc_genes = sc_data.var_names
sc_labels = sc_data.obs.loc[:,"label"]
st_array = st_data.X.toarray()
st_genes = st_data.var_names
x_cord = st_data.obs.array_row
y_cord = st_data.obs.array_col

Perform Deconvolution

deconv_model = sp.SpatialDeconvolution()
cortex_predict = deconv_model.predict_cell_prop(sc_array,st_array,
                                                sc_genes,st_genes,
                                                sc_labels,x_cord,y_cord )

Plot the results

for i in cortex_predict.columns:
    st_data.obs.loc[:,i] = cortex_predict.loc[:,i].values
sc.pl.spatial(st_data,
              color=("L2/3 IT", 'L4', 'L5 IT', 'L5 PT', 'L6 CT', 'L6b',"Oligo"),
              library_id='V1_Mouse_Brain_Sagittal_Anterior')

Spatial domain identification

For domain identification SpatialPrompt requires:

• st_array: Matrix of Spatial data, where rows are the cells and columns are the genes.
• x_cord: X coordinate array of spatial data.
• y_cord: Y coordinate array of spatial data.

Import the library and download the data

import scanpy as sc
import spatialprompt as sp
import gdown
gdown.download("https://drive.google.com/uc?id=1h7E5nPs2ga1ixOBDjLJKK7V8xJq93ez5","st_m_cortex.h5ad", quiet=False)

Load the data into Scanpy and retrieve required information

st_data = sc.read_h5ad("st_m_cortex.h5ad")
#Requires raw data
st_array = st_data.X.toarray()
x_cord = st_data.obs.array_row
y_cord = st_data.obs.array_col

Perform spatial clustering

clus_model = sp.SpatialCluster()
cortex_domains = clus_model.fit_predict(st_array = st_array,
                                        x_cord = x_cord,
                                        y_cord = y_cord,n_cluster=20)
st_data.obs.loc[:,"SpatialPrompt: clusters"] = cortex_domains

Plot the results

sc.pl.spatial(st_data,color=("SpatialPrompt: clusters"),
              library_id='V1_Mouse_Brain_Sagittal_Anterior',
              palette="tab20")