Nature Methods佳作！双注意力机制新型图神经网络SpatialGlue破译空间多组学的空间域

conda create -n SpatialGlue python=3.8conda activate SpatialGluepip install SpatialGlue           ##导入相关的包import osimport torchimport pandas as pdimport scanpy as scimport SpatialGlue           # 环境配置。SpatialGlue包可以用CPU或GPU实现。强烈推荐GPU加速以提高效率。device = torch.device('cuda:2' if torch.cuda.is_available() else 'cpu')           # R的位置，这是“mclust”聚类算法所需要的。请将下面的路径替换为本地R安装路径os.environ['R_HOME'] = '/scbio4/tools/R/R-4.0.3_openblas/R-4.0.3'

# 读取数据file_fold = '/home/yahui/anaconda3/work/SpatialGlue_revision/data/Dataset3_Mouse_Thymus1/' #please replace 'file_fold' with the download path               adata_omics1 = sc.read_h5ad(file_fold + 'adata_RNA.h5ad')adata_omics2 = sc.read_h5ad(file_fold + 'adata_protein.h5ad')           adata_omics1.var_names_make_unique()adata_omics2.var_names_make_unique()           #指定数据类型data_type = 'Stereo-CITE-seq'           #固定随机种子from SpatialGlue.preprocess import fix_seedrandom_seed = 2022fix_seed(random_seed)

from SpatialGlue.preprocess import clr_normalize_each_cell, pca    # RNAsc.pp.filter_genes(adata_omics1, min_cells=10)sc.pp.filter_cells(adata_omics1, min_genes=80)           sc.pp.filter_genes(adata_omics2, min_cells=50)adata_omics2 = adata_omics2[adata_omics1.obs_names].copy()           sc.pp.highly_variable_genes(adata_omics1, flavor="seurat_v3", n_top_genes=3000)sc.pp.normalize_total(adata_omics1, target_sum=1e4)sc.pp.log1p(adata_omics1)           adata_omics1_high =  adata_omics1[:, adata_omics1.var['highly_variable']]adata_omics1.obsm['feat'] = pca(adata_omics1_high, n_comps=adata_omics2.n_vars-1)           # 蛋白质adata_omics2 = clr_normalize_each_cell(adata_omics2)adata_omics2.obsm['feat'] = pca(adata_omics2, n_comps=adata_omics2.n_vars-1)

from SpatialGlue.preprocess import construct_neighbor_graphdata = construct_neighbor_graph(adata_omics1, adata_omics2, datatype=data_type)

# 定义模型from SpatialGlue.SpatialGlue_pyG import Train_SpatialGluemodel = Train_SpatialGlue(data, datatype=data_type, device=device)           # 训练模型output = model.train()           adata = adata_omics1.copy()adata.obsm['emb_latent_omics1'] = output['emb_latent_omics1']adata.obsm['emb_latent_omics2'] = output['emb_latent_omics2']adata.obsm['SpatialGlue'] = output['SpatialGlue']adata.obsm['alpha'] = output['alpha']adata.obsm['alpha_omics1'] = output['alpha_omics1']adata.obsm['alpha_omics2'] = output['alpha_omics2']           6. 跨组学整合分析集成后，使用mclust进行聚类分析。在这里，我们提供了三种可选的聚类工具，包括 mclust、leiden 和 louvain。在我们的实验中，果叔发现在大多数情况下，“mclust”算法在空间数据上的表现优于“leiden”和“louvain”。因此，果叔建议使用“mclust”算法进行聚类。           # 我们将“mclust”默认设置为聚类工具。用户还可以选择“leiden”和“louvain”。tool = 'mclust' # mclust, leiden, and louvainclustering(adata, key='SpatialGlue', add_key='SpatialGlue', n_clusters=8, method=tool, use_pca=True)           #可视化import matplotlib.pyplot as pltadata.obsm['spatial'][:,1] = -1*adata.obsm['spatial'][:,1]           fig, ax_list = plt.subplots(1, 2, figsize=(7, 3))sc.pp.neighbors(adata, use_rep='SpatialGlue', n_neighbors=30)sc.tl.umap(adata)           sc.pl.umap(adata, color='SpatialGlue', ax=ax_list[0], title='SpatialGlue', s=20, show=False)sc.pl.embedding(adata, basis='spatial', color='SpatialGlue', ax=ax_list[1], title='SpatialGlue', s=20, show=False)               plt.tight_layout(w_pad=0.3)plt.show()

# 注释adata.obs['SpatialGlue_number'] = adata.obs['SpatialGlue'].copy()adata.obs['SpatialGlue'].cat.rename_categories({1: '5-Outer cortex region 3(DN T,DP T,cTEC)',                                                2: '7-Subcapsular zone(DN T)',                                                3: '4-Middle cortex region 2(DN T,DP T,cTEC)',                                                4: '2-Corticomedullary Junction(CMJ)',                                                5: '1-Medulla(SP T,mTEC,DC)',                                                6: '6-Connective tissue capsule(fibroblast)',                                                7: '8-Connective tissue capsule(fibroblast,RBC,myeloid)',                                                    8: '3-Inner cortex region 1(DN T,DP T,cTEC)'                                                }, inplace=True)           list_ = ['3-Inner cortex region 1(DN T,DP T,cTEC)','2-Corticomedullary Junction(CMJ)','4-Middle cortex region 2(DN T,DP T,cTEC)',         '7-Subcapsular zone(DN T)', '5-Outer cortex region 3(DN T,DP T,cTEC)', '8-Connective tissue capsule(fibroblast,RBC,myeloid)',         '1-Medulla(SP T,mTEC,DC)','6-Connective tissue capsule(fibroblast)']adata.obs['SpatialGlue']  = pd.Categorical(adata.obs['SpatialGlue'],                      categories=list_,                      ordered=True)# 绘图fig, ax_list = plt.subplots(1, 2, figsize=(9.5, 3))sc.pp.neighbors(adata, use_rep='SpatialGlue', n_neighbors=30)sc.tl.umap(adata)           sc.pl.umap(adata, color='SpatialGlue', ax=ax_list[0], title='SpatialGlue', s=10, show=False)sc.pl.embedding(adata, basis='spatial', color='SpatialGlue', ax=ax_list[1], title='SpatialGlue', s=20, show=False)               ax_list[0].get_legend().remove()           plt.tight_layout(w_pad=0.3)plt.show()

#交换与注释对应的注意权重list_SpatialGlue = [5,4,8,3,1,6,2,7]adata.obs['SpatialGlue_number']  = pd.Categorical(adata.obs['SpatialGlue_number'],                      categories=list_SpatialGlue,                      ordered=True)adata.obs['SpatialGlue_number'].cat.rename_categories({5:1,                                                       4:2,                                                       8:3,                                                       3:4,                                                       1:5,                                                       6:6,                                                       2:7,                                                           7:8                                                }, inplace=True)# 绘制模态权重值import pandas as pdimport seaborn as snsplt.rcParams['figure.figsize'] = (5,3)df = pd.DataFrame(columns=['RNA', 'protein', 'label'])df['RNA'], df['protein'] = adata.obsm['alpha'][:, 0], adata.obsm['alpha'][:, 1]df['label'] = adata.obs['SpatialGlue_number'].valuesdf = df.set_index('label').stack().reset_index()df.columns = ['label_SpatialGlue', 'Modality', 'Weight value']ax = sns.violinplot(data=df, x='label_SpatialGlue', y='Weight value', hue="Modality",                split=True, inner="quart", linewidth=1, show=False)ax.set_title('RNA vs protein')ax.set_xlabel('SpatialGlue label')ax.legend(bbox_to_anchor=(1.4, 1.01), loc='upper right')           plt.tight_layout(w_pad=0.05)#plt.show()