Tutorial 1: Integrating three simulated slices by SMILE

import warnings
warnings.filterwarnings('ignore')

import stSMILE
from stSMILE import SMILE

import scanpy as sc
import anndata as ad
import scanpy.external as sce
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import squidpy as sq
import scipy.sparse as sp
from scipy import sparse
from scipy.sparse import csr_matrix
import math
import torch
import torch.nn as nn
import time
import torch.nn.functional as F
from itertools import chain
from scanpy import read_10x_h5
import torch.optim as optim
import sklearn
from sklearn.neighbors import kneighbors_graph
import gudhi
import networkx as nx
from torch_geometric.nn import GCNConv
import random
import os
import json 
import matplotlib.image as mpimg

Load data

section_ids = ['S1','S2','S3']

adata_l = []
for i in range(len(section_ids)):
    adata_i = sc.read_h5ad('./dataset/simulation/simulated_'+ section_ids[i]+'_ST_final.h5ad')
    adata_i.obs_names = [x+'_'+section_ids[i] for x in adata_i.obs_names]
    cluster = pd.read_csv('/./dataset/simulation/simulated_' + section_ids[i]+'_true_cluster.csv', sep = '\t')
    cluster.index = adata_i.obs_names
    adata_i.obs['Ground Truth'] = pd.Series(cluster['0'])    
    adata_i.obs['Ground Truth'] = adata_i.obs['Ground Truth'].astype(str)
    adata_i.obs['Ground Truth'] = adata_i.obs['Ground Truth'].astype('category')
    adata_l.append(adata_i)

adata0_sc = sc.read_h5ad('./dataset/simulation/simulated_sc_final.h5ad') 
adata_l.append(adata0_sc)

Run SMILE

tag_l = ['ST','ST','ST','single cell']
in_features = len(adata_l[0].var.index)
hidden_features = 512
out_features = 50
feature_method = 'GCNConv'
alpha = 1
beta = 1
gamma = 1000
lamb = 1
theta = 0.9
spatial_regularization_strength= 0.9
lr=1e-3
subepochs=100
epochs=200
max_patience=50
min_stop=20
random_seed=2024
gpu=0
regularization_acceleration=True
edge_subset_sz=1000000
add_topology = True
add_feature = False
add_image = False
add_sc = True
multiscale = True
anchor_type = None
anchors_all = False
use_rep_anchor = 'deconvolution'
align_method = 'MMD'
anchor_size=8000
iter_comb= None
n_clusters_l = [4]
edge_weights = [1,0.1,0.1]

adata_l = SMILE(adata_l, tag_l, section_ids, multiscale,  n_clusters_l, in_features, feature_method, hidden_features, out_features, iter_comb, anchors_all, use_rep_anchor, alpha, beta, lamb, theta, gamma,edge_weights, add_topology, add_feature, add_image, add_sc, spatial_regularization_strength, lr=0.001, subepochs=100, epochs=200,mnn_neigh = 5)

Pretraining to extract embeddings of spots...
epoch   0: train spatial C loss: 0.0000, train F loss: 4.8514,
epoch  10: train spatial C loss: 0.0000, train F loss: 2.2220,
epoch  20: train spatial C loss: 0.0000, train F loss: 1.4567,
epoch  30: train spatial C loss: 0.0000, train F loss: 1.0723,
epoch  40: train spatial C loss: 0.0000, train F loss: 0.8445,
epoch  50: train spatial C loss: 0.0000, train F loss: 0.7221,
epoch  60: train spatial C loss: 0.0000, train F loss: 0.6632,
epoch  70: train spatial C loss: 0.0000, train F loss: 0.6334,
epoch  80: train spatial C loss: 0.0000, train F loss: 0.5898,
epoch  90: train spatial C loss: 0.0000, train F loss: 0.5745,
Training classifier...
Training classifier...
epoch   0: overall loss: 37.6162,sc classifier loss: 1.9628,representation loss: 0.0356,within spatial regularization loss: 0.0980
epoch  10: overall loss: 5.8596,sc classifier loss: 1.2574,representation loss: 0.0045,within spatial regularization loss: 0.0870
epoch  20: overall loss: 3.7902,sc classifier loss: 1.1089,representation loss: 0.0026,within spatial regularization loss: 0.0981
epoch  30: overall loss: 2.2469,sc classifier loss: 0.9047,representation loss: 0.0013,within spatial regularization loss: 0.0979
epoch  40: overall loss: 1.4784,sc classifier loss: 0.6663,representation loss: 0.0007,within spatial regularization loss: 0.0974
epoch  50: overall loss: 1.1395,sc classifier loss: 0.4817,representation loss: 0.0006,within spatial regularization loss: 0.0956
epoch  60: overall loss: 1.0347,sc classifier loss: 0.3506,representation loss: 0.0006,within spatial regularization loss: 0.0947
epoch  70: overall loss: 0.7265,sc classifier loss: 0.2630,representation loss: 0.0004,within spatial regularization loss: 0.0929
epoch  80: overall loss: 0.6014,sc classifier loss: 0.1984,representation loss: 0.0003,within spatial regularization loss: 0.0926
epoch  90: overall loss: 0.5602,sc classifier loss: 0.1562,representation loss: 0.0003,within spatial regularization loss: 0.0920
epoch 100: overall loss: 0.4967,sc classifier loss: 0.1360,representation loss: 0.0003,within spatial regularization loss: 0.0920
epoch 110: overall loss: 0.4817,sc classifier loss: 0.1155,representation loss: 0.0003,within spatial regularization loss: 0.0914
epoch 120: overall loss: 0.4196,sc classifier loss: 0.1053,representation loss: 0.0002,within spatial regularization loss: 0.0910
epoch 130: overall loss: 0.4143,sc classifier loss: 0.0923,representation loss: 0.0002,within spatial regularization loss: 0.0908
epoch 140: overall loss: 0.4955,sc classifier loss: 0.0872,representation loss: 0.0003,within spatial regularization loss: 0.0911
epoch 150: overall loss: 0.3505,sc classifier loss: 0.0793,representation loss: 0.0002,within spatial regularization loss: 0.0906
epoch 160: overall loss: 0.3366,sc classifier loss: 0.0738,representation loss: 0.0002,within spatial regularization loss: 0.0898
epoch 170: overall loss: 0.3546,sc classifier loss: 0.0694,representation loss: 0.0002,within spatial regularization loss: 0.0894
epoch 180: overall loss: 0.4105,sc classifier loss: 0.0654,representation loss: 0.0003,within spatial regularization loss: 0.0897
epoch 190: overall loss: 0.2838,sc classifier loss: 0.0628,representation loss: 0.0001,within spatial regularization loss: 0.0889
single cell data classification: Avg Accuracy = 98.947370%


R[write to console]:                    __           __ 
   ____ ___  _____/ /_  _______/ /_
  / __ `__ \/ ___/ / / / / ___/ __/
 / / / / / / /__/ / /_/ (__  ) /_  
/_/ /_/ /_/\___/_/\__,_/____/\__/   version 6.1.1
Type 'citation("mclust")' for citing this R package in publications.



fitting ...
  |======================================================================| 100%
Identifying anchors...
Processing datasets (0, 1)
Processing datasets (0, 2)
Processing datasets (1, 2)
0.9805634888254228
The ratio of filtered mnn pairs: 1.0
The ratio of filtered mnn pairs: 0.96
The ratio of filtered mnn pairs: 0.9603803486529319
Aligning by anchors...
epoch 100: total loss:4.3614, train F loss: 1.3807, train C loss: 1.3639, train D loss: 2.9807
epoch 110: total loss:1.1600, train F loss: 0.9650, train C loss: 0.3505, train D loss: 0.1950
epoch 120: total loss:1.1445, train F loss: 1.0216, train C loss: 0.1893, train D loss: 0.1229
epoch 130: total loss:0.7761, train F loss: 0.6795, train C loss: 0.0317, train D loss: 0.0966
epoch 140: total loss:0.6860, train F loss: 0.6034, train C loss: 0.0109, train D loss: 0.0826
epoch 150: total loss:0.6461, train F loss: 0.5806, train C loss: 0.0078, train D loss: 0.0656
epoch 160: total loss:0.6199, train F loss: 0.5719, train C loss: 0.0089, train D loss: 0.0479
epoch 170: total loss:0.6012, train F loss: 0.5684, train C loss: 0.0106, train D loss: 0.0328
epoch 180: total loss:0.5889, train F loss: 0.5636, train C loss: 0.0097, train D loss: 0.0253
epoch 190: total loss:0.5809, train F loss: 0.5577, train C loss: 0.0083, train D loss: 0.0232
Updating classifier...
Training classifier...
epoch   0: overall loss: 43.2403,sc classifier loss: 2.0894,representation loss: 0.0411,within spatial regularization loss: 0.1055
epoch  10: overall loss: 6.4191,sc classifier loss: 1.2766,representation loss: 0.0051,within spatial regularization loss: 0.0950
epoch  20: overall loss: 4.9455,sc classifier loss: 1.0837,representation loss: 0.0038,within spatial regularization loss: 0.0946
epoch  30: overall loss: 3.5694,sc classifier loss: 0.9264,representation loss: 0.0026,within spatial regularization loss: 0.0929
epoch  40: overall loss: 2.2210,sc classifier loss: 0.7579,representation loss: 0.0014,within spatial regularization loss: 0.0943
epoch  50: overall loss: 1.4686,sc classifier loss: 0.6345,representation loss: 0.0007,within spatial regularization loss: 0.0941
epoch  60: overall loss: 1.0352,sc classifier loss: 0.5273,representation loss: 0.0004,within spatial regularization loss: 0.0928
epoch  70: overall loss: 0.8816,sc classifier loss: 0.4237,representation loss: 0.0004,within spatial regularization loss: 0.0922
epoch  80: overall loss: 0.7492,sc classifier loss: 0.3363,representation loss: 0.0003,within spatial regularization loss: 0.0918
epoch  90: overall loss: 0.6912,sc classifier loss: 0.2751,representation loss: 0.0003,within spatial regularization loss: 0.0913
epoch 100: overall loss: 0.7203,sc classifier loss: 0.2361,representation loss: 0.0004,within spatial regularization loss: 0.0908
epoch 110: overall loss: 0.6237,sc classifier loss: 0.2100,representation loss: 0.0003,within spatial regularization loss: 0.0906
epoch 120: overall loss: 0.6797,sc classifier loss: 0.1862,representation loss: 0.0004,within spatial regularization loss: 0.0909
epoch 130: overall loss: 0.5683,sc classifier loss: 0.1653,representation loss: 0.0003,within spatial regularization loss: 0.0911
epoch 140: overall loss: 0.4838,sc classifier loss: 0.1461,representation loss: 0.0003,within spatial regularization loss: 0.0903
epoch 150: overall loss: 0.4287,sc classifier loss: 0.1301,representation loss: 0.0002,within spatial regularization loss: 0.0899
epoch 160: overall loss: 0.4165,sc classifier loss: 0.1177,representation loss: 0.0002,within spatial regularization loss: 0.0899
epoch 170: overall loss: 0.7066,sc classifier loss: 0.1077,representation loss: 0.0005,within spatial regularization loss: 0.0903
epoch 180: overall loss: 0.4558,sc classifier loss: 0.0981,representation loss: 0.0003,within spatial regularization loss: 0.0903
epoch 190: overall loss: 0.3739,sc classifier loss: 0.0892,representation loss: 0.0002,within spatial regularization loss: 0.0896
single cell data classification: Avg Accuracy = 98.693955%

adata_concat_st = ad.concat(adata_l[0:len(section_ids)], label="slice_name", keys=section_ids)

adata_concat_st.obs['pd_cluster'] = adata_concat_st.obs['pd_cluster'].astype(str)
adata_concat_st.obs['pd_cluster'] = adata_concat_st.obs['pd_cluster'].astype('category')

sc.tl.pca(adata_concat_st)
adata_concat_st.obsm['X_pca_old'] = adata_concat_st.obsm['X_pca'].copy()
adata_concat_st.obsm['X_pca'] = adata_concat_st.obsm['embedding'].copy()
sc.pp.neighbors(adata_concat_st)  
sc.tl.umap(adata_concat_st)

Results and visualizations

import matplotlib.pyplot as plt
from sklearn.metrics import adjusted_rand_score as ari_score
from sklearn.metrics import normalized_mutual_info_score as nmi_score
import seaborn as sns
rgb_values = sns.color_palette("Set1", len(adata_concat_st.obs['pd_cluster'].unique()))
color_fine = dict(zip(list(adata_concat_st.obs['pd_cluster'].unique()), rgb_values))
Batch_list = []
for section_id in section_ids:
    Batch_list.append(adata_concat_st[adata_concat_st.obs['slice_name'] == section_id])

spot_size = 0.8
title_size = 12
ARI_list = []
NMI_list = []
for bb in range(len(section_ids)):
    ARI_list.append(round(ari_score(Batch_list[bb].obs['Ground Truth'], Batch_list[bb].obs['pd_cluster']), 2))
    NMI_list.append(round(nmi_score(Batch_list[bb].obs['Ground Truth'], Batch_list[bb].obs['pd_cluster']), 2))

fig, ax = plt.subplots(3,1, figsize=(3.5,7), gridspec_kw={'wspace': 0.05, 'hspace': 0.1})
_sc_0 = sc.pl.spatial(Batch_list[0], img_key=None, color=['pd_cluster'], 
                      legend_loc='right margin', legend_fontsize=12, show=False, ax=ax[0], frameon=False,
                      spot_size=spot_size,palette = color_fine)
_sc_0[0].set_title("ARI=" + str(ARI_list[0])+",NMI=" + str(NMI_list[0]), size=title_size)
_sc_1 = sc.pl.spatial(Batch_list[1], img_key=None, color=['pd_cluster'], 
                      legend_loc='right margin', legend_fontsize=12, show=False, ax=ax[1], frameon=False,
                      spot_size=spot_size,palette = color_fine)
_sc_1[0].set_title("ARI=" + str(ARI_list[1])+",NMI=" + str(NMI_list[1]), size=title_size)

_sc_2 = sc.pl.spatial(Batch_list[2], img_key=None, color=['pd_cluster'], 
                      legend_loc='right margin', legend_fontsize=12, show=False, ax=ax[2], frameon=False,
                      spot_size=spot_size,palette = color_fine)
_sc_2[0].set_title("ARI=" + str(ARI_list[2])+",NMI=" + str(NMI_list[2]), size=title_size)

plt.savefig("simulated_spatial_SMILE.pdf") 
plt.show()