Spatiotemporal pattern

The Spatiotemporal Pattern function in SOAPy employs tensor decomposition to extract components from the three-order expression tensor (“Time-Space-Gene”), revealing hidden patterns and reducing the complexity of data explanation. We used Nanosting DSP mouse developmental samples to explore the spatio-temporal developmental relationships of three major organs: heart, lung, and midgut.

Download ‘Count Results’ and ‘Images’ files from https://nanostring.com/products/geomx-digital-spatial-profiler/spatial-organ-atlas/mouse-development/.

Read and preprocessing

import warnings
warnings.filterwarnings("ignore")
import SOAPy_st as sp
import scanpy as sc
import numpy as np

adata = sp.pp.read_dsp2adata(
    xml_file={
        'mu_dev_E9_001': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E9_image_files/mu_dev_E9_001.ome.xml',
        # The reason is that the keywords in the xml file of mu_dev_E9_002 are different from other .xml files.
        # If you want to read the mu_dev_E9_002 .xml file together, you need to change the "Polyline" to "Polygon" in the mu_dev_E9_002.ome.xml file.
        'mu_dev_E9_003': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E9_image_files/mu_dev_E9_003.ome.xml',
        'mu_dev_E11_004': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E11_image_files/mu_dev_E11_004.ome.xml',
        'mu_dev_E11_005': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E11_image_files/mu_dev_E11_005.ome.xml',
        'mu_dev_E13_006': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E13_image_files/mu_dev_E13_006.ome.xml',
        'mu_dev_E13_007': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E13_image_files/mu_dev_E13_007.ome.xml',
        'mu_dev_E13_008': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E13_image_files/mu_dev_E13_008.ome.xml',
        'mu_dev_E13_009': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E13_image_files/mu_dev_E13_009.ome.xml',
        'mu_dev_E13_010': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E13_image_files/mu_dev_E13_010.ome.xml',
        'mu_dev_E13_011': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E13_image_files/mu_dev_E13_011.ome.xml',
        'mu_dev_E15_012': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E15_image_files/mu_dev_E15_012.ome.xml',
        'mu_dev_E15_013': '/csb2/project/tmp_download/GeoMx_MouseDevelopment/image_files/mu_dev_E15_image_files/mu_dev_E15_013.ome.xml'
    },
    information_file='/csb2/project/SpatialPackage_whq/Tutorial/data/nanostring_growth/Export4_NormalizationQ3.xlsx',
)

Firstly, the AOI of the three organs were extracted in adata.

adata = adata[np.isin(adata.obs['Tissue'], ['Heart', 'Lung', 'Midgut']), :].copy()

sc.pp.log1p(adata)
# sc.pp.highly_variable_genes(adata, flavor="seurat", n_top_genes=1000)
# adata = adata[:, adata.var['highly_variable'] == True].copy()

adata.obs['Tissue_s'] = adata.obs['Tissue'].astype(str) + ' ' + adata.obs['Tissue_substructure'].astype(str)

Tensor decomposition

In the following, 3D-tensor is constructed by subtissues, time points, and genes.

Tensors can be generated automatically by sp.tl.TensorDecomposition() and contain gene information by default, only requiring the user to provide labels for other dimensions.

tensor = sp.tl.TensorDecomposition()
tensor.tensor_with_gene(adata, obs_factor=['Tissue_s', 'Timepoint'])

The first 1000 hypervariable genes are selected and standardized, and then the nonnegative CANDECOMP/PARAFAC(CP) tensor with rank 7 is decomposed by Tensor.cp_decomposition() to obtain the factor matrix and weight of the 7 factors.

tensor.highly_variable(factor_name='gene', top_num=1000) # select 1000 high variable genes
tensor.normalization(factor_name='gene', method='max') # normalization used the maximum value
weights, factors, nre = tensor.CP_decomposition(rank=7, non_negative=True, random_state=10, use_hals=True)

Visualizing the results

Three bars show the composition of subtissues, time points, and genes within each factor.

sp.pl.show_proportion_in_CP(tensor, module=0, max_key=10)

sp.pl.show_proportion_in_CP(tensor, module=4, max_key=10)