Splane tutorial: Identification of uniform spatial domain on human breast cancer Visium dataset
July 2023
Dataset: 11 Visium slices of human breast cancer (here).
[1]:
from SPACEL.setting import set_environ_seed
set_environ_seed(42)
from SPACEL import Splane
import scanpy as sc
import matplotlib
import pandas as pd
matplotlib.rcParams['pdf.fonttype'] = 42
matplotlib.rcParams['ps.fonttype'] = 42
matplotlib.rcParams['font.serif'] = ['Arial']
sc.settings.set_figure_params(dpi=100,dpi_save=300,facecolor='white',fontsize=10,vector_friendly=True,figsize=(3,3))
Setting environment seed: 42
Spatial transcriptomic data import
The input data type of Splane is a list of anndata objects for each slice from Spoint. If users use cell type composition from other deconvoluiton methods, the anndata objects of each slice must contain cell type composition in .obs and cell types name in .uns['celltypes']
[2]:
st_1142243F_ad = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_1142243F.h5ad')
st_1160920F_ad = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_1160920F.h5ad')
st_CID4290_ad = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_CID4290.h5ad')
st_CID4465_ad = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_CID4465.h5ad')
st_CID44971_ad = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_CID44971.h5ad')
st_CID4535_ad = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_CID4535.h5ad')
Parent_Visium_Human_BreastCancer = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_Parent_Visium_Human_BreastCancer.h5ad')
V1_Breast_Cancer_Block_A_Section_1 = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_V1_Breast_Cancer_Block_A_Section_1.h5ad')
V1_Breast_Cancer_Block_A_Section_2 = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_V1_Breast_Cancer_Block_A_Section_2.h5ad')
V1_Human_Invasive_Ductal_Carcinoma = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_V1_Human_Invasive_Ductal_Carcinoma.h5ad')
Visium_FFPE_Human_Breast_Cancer = sc.read_h5ad('../data/visium_human_breast_cancer/human_bc_spatial_Visium_FFPE_Human_Breast_Cancer.h5ad')
Loading the cell type composition of spots of these slices predicted by Spoint.
[3]:
st_1142243F_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_1142243F.csv',index_col=0)
st_1160920F_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_1160920F.csv',index_col=0)
st_CID4290_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_CID4290.csv',index_col=0)
st_CID4465_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_CID4465.csv',index_col=0)
st_CID44971_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_CID44971.csv',index_col=0)
st_CID4535_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_CID4535.csv',index_col=0)
Parent_Visium_Human_BreastCancer_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_Parent_Visium_Human_BreastCancer.csv',index_col=0)
V1_Breast_Cancer_Block_A_Section_1_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_V1_Breast_Cancer_Block_A_Section_1.csv',index_col=0)
V1_Breast_Cancer_Block_A_Section_2_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_V1_Breast_Cancer_Block_A_Section_2.csv',index_col=0)
V1_Human_Invasive_Ductal_Carcinoma_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_V1_Human_Invasive_Ductal_Carcinoma.csv',index_col=0)
Visium_FFPE_Human_Breast_Cancer_prop = pd.read_csv('../data/visium_human_breast_cancer/human_bc_spatial_Visium_FFPE_Human_Breast_Cancer.csv',index_col=0)
Here, we will incorporate the cell type composition predicted by Spoint into the spatial anndata object for subsequent spatial domain identification in Splane using the add_cell_type_composition function. This function takes a DataFrame containing the cell type composition matrix as input for spot-based spatial transcriptomic data or a series of cell type annotations as input for single-cell resolution spatial transcriptomic data.
[ ]:
Splane.utils.add_cell_type_composition(st_1142243F_ad,st_1142243F_prop)
Splane.utils.add_cell_type_composition(st_1160920F_ad,st_1160920F_prop)
Splane.utils.add_cell_type_composition(st_CID4290_ad,st_CID4290_prop)
Splane.utils.add_cell_type_composition(st_CID4465_ad,st_CID4465_prop)
Splane.utils.add_cell_type_composition(st_CID44971_ad,st_CID44971_prop)
Splane.utils.add_cell_type_composition(st_CID4535_ad,st_CID4535_prop)
Splane.utils.add_cell_type_composition(Parent_Visium_Human_BreastCancer,Parent_Visium_Human_BreastCancer_prop)
Splane.utils.add_cell_type_composition(V1_Breast_Cancer_Block_A_Section_1,V1_Breast_Cancer_Block_A_Section_1_prop)
Splane.utils.add_cell_type_composition(V1_Breast_Cancer_Block_A_Section_2,V1_Breast_Cancer_Block_A_Section_2_prop)
Splane.utils.add_cell_type_composition(V1_Human_Invasive_Ductal_Carcinoma,V1_Human_Invasive_Ductal_Carcinoma_prop)
Splane.utils.add_cell_type_composition(Visium_FFPE_Human_Breast_Cancer,Visium_FFPE_Human_Breast_Cancer_prop)
Concatting spatial anndata objects to a single list as a input for Splane
[5]:
st_ad_list = [
st_1142243F_ad,
st_1160920F_ad,
st_CID4290_ad,
st_CID4465_ad,
st_CID44971_ad,
st_CID4535_ad,
Parent_Visium_Human_BreastCancer,
V1_Breast_Cancer_Block_A_Section_1,
V1_Breast_Cancer_Block_A_Section_2,
V1_Human_Invasive_Ductal_Carcinoma,
Visium_FFPE_Human_Breast_Cancer
]
Training Splane model
In this step, we initialize the Splane model using the anndata object list as input. The n_clusters parameter determines the number of spatial domains to be identified. The k parameter controls the degree of neighbors considered in the model, with a larger k value resulting in more emphasis on global structure rather than local structure. The gnn_dropout parameter influences the level of smoothness in the model’s predictions, with a higher gnn_dropout value resulting in a
smoother output that accommodates the sparsity of the spatial transcriptomics data.
[8]:
splane_model = Splane.init_model(st_ad_list, n_clusters=10, k=2, gnn_dropout=0.8, use_gpu=False)
Setting global seed: 42
Calculating cell type weights...
Generating GNN inputs...
Calculating largest eigenvalue of normalized graph Laplacian...
Calculating Chebyshev polynomials up to order 2...
Here, we train the model to obtain latent feature of each spots/cells. The parameter d_l affects the level of batch effect correction between slices. By default, d_l is 0.5 for spot-based spatial transcriptomics data.
[9]:
splane_model.train(d_l=0.5)
The best epoch 95 total loss=-4.384 g loss=-3.189 d loss=2.389 d acc=0.153 simi loss=-0.993 db loss=0.701: 50%|█████ | 150/300 [22:13<22:13, 8.89s/it]
Stop trainning because of loss convergence
Then, we can identify the spatial domain to which each spot/cell belongs. By default, the results will be saved in spatial_domain column in .obs. If the key parameter is provided, the results will be saved in .obs[key].
[10]:
splane_model.identify_spatial_domain()
Plot spatial domain results
[11]:
for ad in st_ad_list:
sc.pl.spatial(ad,color='spatial_domain')
