SpatialScope

https://github.com/yanglabhkust/spatialscope
Admin
SpatialScope is a unified approach for integrating spatial and single-cell transcriptomics data by...
Tokens:150,191
Snippets:810
Trust Score:9.1
Update:4 months ago
Show doc for...
Tokens:
Raw
### Check Installation Status

Source: https://github.com/yanglabhkust/spatialscope/blob/master/README.md

Verifies the installation by running a Python script with the help flag.

```shell
python ./src/Cell_Type_Identification.py -h
```

--------------------------------

### Plotting Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Initializes plotting parameters and potentially sets up axes for visualization.

```python
ax.axis('off')
ax[1, 1].set_title('error rate: {:.4f}'.format((SR_cell_type_label_vis.error == 1).sum() / SR_cell_type_label_vis.shape[0]))
ax[1, 1].get_yaxis().set_visible(False)
ax[1, 1].get_xaxis().set_visible(False)
ax[1, 1].get_legend().remove()
for i in range(x_seg.shape[0]):
    ax[1, 1].plot([x_seg[i] for _ in y_seg],y_seg,linestyle='dashed', c = 'salmon', lw = .2, alpha = 0.5)
for i in range(y_seg.shape[0]):    
    ax[1, 1].plot(x_seg,[y_seg[i] for _ in x_seg],linestyle='dashed', c = 'salmon', lw = .2, alpha = 0.5)
```

--------------------------------

### Plotting Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Initializes plot settings and creates a figure and axes for plotting.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(1,1, figsize = (20, 7), dpi = 150)
plt.subplots_adjust(hspace = 0.5)
```

--------------------------------

### Plotting Setup (3 Methods)

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Initializes plot settings and creates a figure and axes for plotting when comparing three methods.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(1,1, figsize = (28, 7), dpi = 150)
plt.subplots_adjust(hspace = 0.5)
```

--------------------------------

### Plotting setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Sets up matplotlib and seaborn for plotting, including style and context.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(1,1, figsize = (20, 8), dpi = 150)
plt.subplots_adjust(hspace = 0.5)

with plt.style.context('seaborn-whitegrid'):

```

--------------------------------

### Directory Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_6.ipynb

Sets up base directories for saving data based on grid dimensions and sample UMI counts.

```python
base_path = '/home/share/xwanaf/sour_sep/revision/data/RCTD_MOp_3D'
if not os.path.exists(base_path):
    os.mkdir(base_path)
base_path = os.path.join(base_path, str(num_x) + 'x' + str(num_y))
if not os.path.exists(base_path):
    os.mkdir(base_path)

for sample_UMI in sample_UMIs:
    save_data_dir = os.path.join(base_path, f'{int(sample_UMI)}UMI')
    print(save_data_dir)
    if not os.path.exists(save_data_dir):
        os.mkdir(save_data_dir)
```

--------------------------------

### Installation

Source: https://github.com/yanglabhkust/spatialscope/blob/master/README.md

Clones the repository, sets up a conda environment, and applies a fix for squidpy.

```shell
git clone https://github.com/YangLabHKUST/SpatialScope.git
cd SpatialScope
conda env create -f environment.yml
conda activate SpatialScope
# fix bug of squidpy, locate the lib with `which python`
rsync ./src/_feature_mixin.py ~/.conda/envs/SpatialScope/lib/python3.9/site-packages/squidpy/im/_feature_mixin.py
```

--------------------------------

### Plotting Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Configures matplotlib and seaborn for plotting, setting style, context, figure size, and DPI.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=4) 
fig, ax = plt.subplots(1,1, figsize = (40, 8), dpi = 150)
plt.subplots_adjust(hspace = 0.5)

palette = ['#EB8E83'] * (len(vis_nu_list) - 1) + ['#D8A8EB']
with plt.style.context('seaborn-whitegrid'):

```

--------------------------------

### Legend Example

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_5.ipynb

Example of creating a legend for a plot.

```python
legend2 = ax.legend(handles, labels, loc="upper right", title="cell number")
```

--------------------------------

### Plotting Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_6.ipynb

This snippet sets up the plotting environment using matplotlib and seaborn, including style, context, figure size, and color palette.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid') # seaborn-whitegrid

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(2,1, figsize = (20, 7), dpi = 150, gridspec_kw={'height_ratios': [1, 3]})

starting_point = 0
bar_width = 3 #1.5
width_between_bar = 2 #1
order = ['single slice', 'multiple slices']
hue_len = len(order) 
dist = bar_width * hue_len + width_between_bar
    
palette = ['#EB8E83', '#A02439'] #* (len(vis_nu_list) - 1) + ['#D8A8EB']
```

--------------------------------

### RMSE Plot Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_3.ipynb

Sets up plotting parameters and displays a bar plot of RMSE values for different methods and nu values.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(1,1, figsize = (20, 7), dpi = 150)
plt.subplots_adjust(hspace = 0.5)

with plt.style.context('seaborn-whitegrid'):
    sns.barplot(x="nu", y="value",hue='method',data=vis_decon_df[vis_decon_df.metric == 'rmse'], palette=['#EA8E83', '#AFCF78', '#8AAEC9'], ax = ax)
# ax.set_ylim(0.4,0.95)
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
plt.legend(bbox_to_anchor=(1,0.6))
#ax.set_yticks([0.84, 0.86, 0.88])
#ax.set_xticklabels(['130 UMIs','260 UMIs','520 UMIs'])
plt.ylabel('RMSE')
plt.xlabel(r'$
u$ value')
plt.show()
```

--------------------------------

### PCC Plot Setup

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_3.ipynb

Sets up plotting parameters and displays a bar plot of PCC values for different methods and nu values.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(1,1, figsize = (20, 7), dpi = 150)
plt.subplots_adjust(hspace = 0.5)

with plt.style.context('seaborn-whitegrid'):
    sns.barplot(x="nu", y="value",hue='method',data=vis_decon_df[vis_decon_df.metric == 'pcc'], palette=['#EA8E83', '#AFCF78', '#8AAEC9'], ax = ax)
ax.set_ylim(0.7,1.02)
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
plt.legend(bbox_to_anchor=(1,0.6))
#ax.set_yticks([0.84, 0.86, 0.88])
#ax.set_xticklabels(['130 UMIs','260 UMIs','520 UMIs'])
plt.ylabel('PCC')
plt.xlabel(r'$
u$ value')
plt.show()
```

--------------------------------

### Scatter Plot Example

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_4.ipynb

Example of creating a scatter plot using seaborn.

```python
ax = sns.scatterplot(data=sp_adata.obs, x="X", y="Y", s = 50, palette = "rocket_r")
```

--------------------------------

### Import Libraries

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-MOp-MERFISH.ipynb

Imports necessary Python libraries for data analysis and visualization.

```python
import numpy as np
import pandas as pd
import pathlib
import matplotlib.pyplot as plt
import matplotlib as mpl
import seaborn as sns
import scanpy as sc
import pickle

import sys
sys.path.append('../src')
from utils import *

import warnings
warnings.filterwarnings('ignore')
```

--------------------------------

### Set up base path and create directories

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_5.ipynb

Sets the base path for saving data and creates necessary directories for different sample UMI counts.

```python
base_path = '/home/share/xwanaf/sour_sep/revision/data/RCTDnv_STARMAP_PLUS'
if not os.path.exists(base_path):
    os.mkdir(base_path)
base_path = os.path.join(base_path, str(num_x) + 'x' + str(num_y))
if not os.path.exists(base_path):
    os.mkdir(base_path)

for sample_UMI in sample_UMIs:
    save_data_dir = os.path.join(base_path, f'{int(sample_UMI)}UMI')
    print(save_data_dir)
    if not os.path.exists(save_data_dir):
        os.mkdir(save_data_dir)
```

--------------------------------

### Get shape of spatial data

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Cerebellum-Slideseq.ipynb

Prints the shape (number of cells and genes) of the spatial data object.

```python
ad_sp.shape
```

--------------------------------

### Load Initial Proportions

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Cerebellum-Slideseq.ipynb

Loads initial proportion data from a pickle file.

```python
with open('../output/cere/InitProp.pickle', 'rb') as handle:
    InitProp = pickle.load(handle)
```

--------------------------------

### Loading Results

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Loads benchmarking results from specified directories.

```bash
load results from /home/share/xwanaf/sour_sep/revision/data/benchmarking_tutorial/130UMI
```

```bash
load results from /home/share/xwanaf/sour_sep/revision/data/benchmarking_tutorial/260UMI
```

```bash
load results from /home/share/xwanaf/sour_sep/revision/data/benchmarking_tutorial/520UMI
```

--------------------------------

### Get Unique Spatial Cell Types

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Retrieves and prints the unique cell types from the spatial data.

```python
np.unique(sp_adata.obs[sp_cell_class_column])
```

--------------------------------

### Training scRNA-seq reference model

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Command to train a scRNA-seq reference model using SpatialScope.

```bash
python ./src/Train_scRef.py \
--ckpt_path ./Ckpts_scRefs/VISp \
--scRef ./Ckpts_scRefs/VISp/Ref_scRNA_VISp_qc2_2Kgenes.h5ad \
--cell_class_column cell_subclass \
--gpus 0,1,2,3 
```

--------------------------------

### Get Unique Cell Types

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Retrieves and prints the unique cell types from the single-cell reference data.

```python
np.unique(sc_adata.obs[cell_class_column])
```

--------------------------------

### Run SpatialScope for Gene Expression Decomposition (Part 1)

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Command to run SpatialScope for gene expression decomposition for the first spot range.

```bash
python ./src/Decomposition.py --tissue cortex --out_dir  ./output --SC_Data ./Ckpts_scRefs/VISp/Ref_scRNA_VISp_qc2_2Kgenes.h5ad --cell_class_column cell_subclass  --ckpt_path ./Ckpts_scRefs/VISp/model_5000.pt --spot_range 0,500 --gpu 2,3,4,8
```

--------------------------------

### Figure Setup for Multiple Plots

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_5.ipynb

Initializes a matplotlib figure and axes for plotting, with specific height ratios for subplots.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(2,1, figsize = (20, 7), dpi = 150, gridspec_kw={'height_ratios': [4, 1]})

with plt.style.context('seaborn-whitegrid'):
    
```

--------------------------------

### Visualize the data

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Sets up plotting context and generates scatter plots for both simulated spatial data and simulated single-cell reference data, highlighting cell types.

```python
sns.set_context('paper',font_scale=3) 
plt.rcParams["legend.markerscale"] = 3

plt.figure(figsize=(24, 12),dpi = 100)
ax1 = plt.subplot(4,6,(8,14))
ax2 = plt.subplot(1,2,2)
ax = [ax1, ax2]

sns.scatterplot(data=sp_adata.obs, x="x", y="y", hue='cell_type_annot', s = 15, palette = color_dict, ax = ax[0], legend = False)
ax[0].set_title('data used to make \n simulated spatial data')
ax[0].invert_yaxis()
ax[0].axis('off')
# ax[0].legend(bbox_to_anchor=(1.02, 1), loc='upper left', borderaxespad=0, frameon=False)
sns.scatterplot(data=sc_adata.obs, x="x", y="y", hue='cell_type_annot', s = 15, palette = color_dict, ax = ax[1])
ax[1].set_title('simulated single-cell \n reference data')
ax[1].invert_yaxis()
ax[1].axis('off')
ax[1].legend(bbox_to_anchor=(1.02, 1), loc='upper left', borderaxespad=0, frameon=False)
```

--------------------------------

### Get Unique Cell Types

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_4.ipynb

Retrieves and prints the unique cell type annotations present in the spatial data.

```python
np.unique(sp_adata.obs['cell_type_annot'])
```

--------------------------------

### Get Unique Values of 'nu'

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_5.ipynb

Retrieves and displays the unique values present in the 'nu' column of the error rate DataFrame.

```python
np.unique(error_rate_df['nu'])
```

--------------------------------

### Run SpatialScope for Gene Expression Decomposition (Part 2)

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Command to run SpatialScope for gene expression decomposition for the second spot range.

```bash
python ./src/Decomposition.py --tissue cortex --out_dir  ./output --SC_Data ./Ckpts_scRefs/VISp/Ref_scRNA_VISp_qc2_2Kgenes.h5ad --cell_class_column cell_subclass  --ckpt_path ./Ckpts_scRefs/VISp/model_5000.pt --spot_range 500,1000 --gpu 2,3,4,8
```

--------------------------------

### Get unique values of 'nu' column

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_4.ipynb

Finds and returns the unique values present in the 'nu' column of the decon_df DataFrame.

```python
np.unique(decon_df.nu)
```

--------------------------------

### Data Loading and Grid Initialization

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Loads spatial and single-cell reference data, defines cell class column, and sets up grid parameters.

```python
sp_adata = sc.read('/home/xwanaf/bio/multiSlices_SS/mouseOB/SpatialScope/demo_data/revision/benchmarking/MERFISH_MBA_sp.h5ad')
sc_adata = sc.read('/home/xwanaf/bio/multiSlices_SS/mouseOB/SpatialScope/demo_data/revision/benchmarking/MERFISH_MBA_scref.h5ad')

sp_cell_class_column = cell_class_column = 'cell_type_annot'

x_min = sp_adata.obs['x'].min()
x_max = sp_adata.obs['x'].max()
y_min = sp_adata.obs['y'].min()
y_max = sp_adata.obs['y'].max()

num_x = 48
num_y = 48

spatial_adata =  sp_adata.copy()

# make grid 
x_seg = np.linspace(x_min, x_max, num_x)
y_seg = np.linspace(y_min, y_max, num_y)

```

--------------------------------

### Train scRNA-seq reference

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-MOp-MERFISH.ipynb

This command trains a scRNA-seq reference model using four GPUs. It specifies the checkpoint path, the reference data, the cell class column, and the GPUs to use.

```bash
python ./src/Train_scRef.py \
--ckpt_path ./Ckpts_scRefs/MOp \
--scRef ./Ckpts_scRefs/MOp/Ref_snRNA_mop_qc3_2Kgenes.h5ad \
--cell_class_column subclass_label \
--gpus 0,1,2,3 
```

--------------------------------

### Data Preparation for Visualization (3 Methods)

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Prepares a DataFrame for visualizing results including RCTD, SpatialScope, and StarDist+RCTD.

```python
vis_decon_df = pd.DataFrame()
vis_decon_df['value'] = decon_df_temp.RCTD.tolist() + decon_df_temp.SpatialScope.tolist() + decon_df_temp.SR.tolist()
vis_decon_df['method'] = ['RCTD'] * decon_df_temp.shape[0] + ['SpatialScope'] * decon_df_temp.shape[0] + ['StarDist+RCTD'] * decon_df_temp.shape[0]
vis_decon_df['nu'] = decon_df.nu.tolist() * 3
vis_decon_df['metric'] = decon_df.metric.tolist() * 3
```

--------------------------------

### Run SpatialScope for Gene Expression Decomposition (Part 3)

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Command to run SpatialScope for gene expression decomposition for the third spot range.

```bash
python ./src/Decomposition.py --tissue cortex --out_dir  ./output --SC_Data ./Ckpts_scRefs/VISp/Ref_scRNA_VISp_qc2_2Kgenes.h5ad --cell_class_column cell_subclass  --ckpt_path ./Ckpts_scRefs/VISp/model_5000.pt --spot_range 1000,1715 --gpu 1,2,4,8
```

--------------------------------

### Iterating through Deconvolution Parameters

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_5.ipynb

Python code snippet demonstrating the setup for iterating through different deconvolution parameters (nu) and repetitions (rep), including loading initial proportions.

```python
nu_list = [0, 0.000001, 5] + np.arange(10,110,10).tolist() + [120, 150, 200, 500, 1000] + ['RCTD']

decon_df = pd.DataFrame()
decon_pcc_df = pd.DataFrame()
decon_rmse_df = pd.DataFrame()

for j, nu in enumerate(nu_list):
    for i, rep in enumerate(range(10)):
        print(f"nu = {nu}; rep = {rep}")
        
        ################################################################################################################## RCTD resutls
        with open(os.path.join(DataDir, 'rep' + str(rep), 'InitProp.pickle'), 'rb') as handle:
            InitProp = pickle.load(handle)
```

--------------------------------

### Prepare Plotting DataFrame and Setup Plotting Environment

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Creates a DataFrame 'df_plot' from the UMAP embeddings, assigns sizes and color categories based on cell type (reference, selected spatial, generated), and sets up the plotting style and parameters for a high-resolution figure.

```python
df_plot = pd.DataFrame(adata_all.obsm['X_umap'], columns = ['x', 'y'])
color_mu = np.ones(df_plot.shape[0])
color_mu[-(n_cells + n_cells_ss):-(n_cells_ss)] = 12
color_mu[-(n_cells_ss):] = 4
df_plot['size'] = color_mu
color_mu = np.array(['All cells from the single-cell reference data                                ' for _ in range(df_plot.shape[0])])
color_mu[-(n_cells + n_cells_ss):-(n_cells_ss)] = 'Selected scRNA-seq Cells from spatial'
color_mu[-(n_cells_ss):] = 'Generated Cells by SpatialScope'
df_plot['color'] = color_mu


# sns.set_context('paper',font_scale=3) 
# plt.rcParams["legend.markerscale"] = 3

sns.set_style('white')
sns.set_context('paper',font_scale=2.6)
plt.rcParams["legend.markerscale"] = 2.6
colors = ['#C3D5CB','#2ca02c','#d62728','#9467bd','#ff7f0e']
fig, ax = plt.subplots(1,1,figsize = (12,12),dpi = 150)
```

--------------------------------

### Initialization

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_6.ipynb

Initializes lists and dataframes for storing results.

```python
nu_list = [0, 5] + np.arange(10,110,10).tolist() + [120, 150, 200, 500, 1000] + ['RCTD']

error_rate_df = pd.DataFrame(columns = {'error_rate', 'nu', '3D', 'rep'})
decon_df = pd.DataFrame()
decon_pcc_df = pd.DataFrame()
decon_rmse_df = pd.DataFrame()
```

--------------------------------

### Plotting Setup and Error Rate Bar Plot

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Sets up the plotting environment using seaborn and matplotlib, then generates a bar plot for error rates. This includes setting the figure size, style, and context, and displaying a horizontal line for the mean RCTD error rate.

```python
plt.rcParams.update(plt.rcParamsDefault)
%matplotlib inline
plt.style.use('seaborn-whitegrid')

sns.set_context('paper',font_scale=2) 
fig, ax = plt.subplots(1,1, figsize = (24, 8), dpi = 150)
plt.subplots_adjust(hspace = 0.5)

with plt.style.context('seaborn-whitegrid'):
#     sns.barplot(x="nu", y="error_rate",hue='subsample',data=vis_df[(vis_df['correct_label'])&(vis_df['cell_num']==1)], palette="Set3", ax = ax)
    sns.barplot(x="nu", y="error_rate",data=error_rate_df, palette="Set3", ax = ax)
ax.set_ylim(0.17,0.32)
ax.spines['right'].set_visible(False)
ax.spines['top'].set_visible(False)
ax.set_xticklabels(vis_nu_list)

plt.xlabel(r'$\nu$ value & StarDist + RCTD')
plt.axhline(y=error_rate_df[error_rate_df['nu'] == 'RCTD'].error_rate.mean(), color='r', linestyle='--')
plt.show()
```

--------------------------------

### Load and preprocess Visium data for slice 2

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Loads the Visium data for slice 2, applies a transformation to the expression matrix (exponential of log1p), and prepares it for analysis.

```python
slice2_file = '../demo_data/Visium_MouceBrain_Cortex_section2.h5ad'
slice2 = sc.read(slice2_file)
slice2.X = np.exp(slice2.X.A)-1
```

--------------------------------

### Import Libraries

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Cerebellum-Slideseq.ipynb

Imports necessary libraries for data analysis and visualization.

```python
import numpy as np
import pandas as pd
import pathlib
import matplotlib.pyplot as plt
import matplotlib as mpl
import seaborn as sns
import scanpy as sc

import sys
sys.path.append('../src')
from utils import *
import pickle

import warnings
warnings.filterwarnings('ignore')
```

--------------------------------

### Parameter Iteration

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Iterates through different values of 'nu' and 'rep' for benchmarking.

```text
nu = 0; rep = 0
nu = 0; rep = 1
nu = 0; rep = 2
nu = 0; rep = 3
nu = 0; rep = 4
nu = 0; rep = 5
nu = 0; rep = 6
nu = 0; rep = 7
nu = 0; rep = 8
nu = 0; rep = 9
nu = 5; rep = 0
nu = 5; rep = 1
nu = 5; rep = 2
nu = 5; rep = 3
nu = 5; rep = 4
nu = 5; rep = 5
nu = 5; rep = 6
nu = 5; rep = 7
nu = 5; rep = 8
nu = 5; rep = 9
nu = 10; rep = 0
nu = 10; rep = 1
nu = 10; rep = 2
nu = 10; rep = 3
nu = 10; rep = 4
nu = 10; rep = 5
nu = 10; rep = 6
nu = 10; rep = 7
nu = 10; rep = 8
nu = 10; rep = 9
nu = 20; rep = 0
nu = 20; rep = 1
nu = 20; rep = 2
nu = 20; rep = 3
nu = 20; rep = 4
nu = 20; rep = 5
nu = 20; rep = 6
nu = 20; rep = 7
nu = 20; rep = 8
nu = 20; rep = 9
nu = 30; rep = 0
nu = 30; rep = 1
nu = 30; rep = 2
nu = 30; rep = 3
nu = 30; rep = 4
nu = 30; rep = 5
nu = 30; rep = 6
nu = 30; rep = 7
nu = 30; rep = 8
nu = 30; rep = 9
nu = 40; rep = 0
nu = 40; rep = 1
nu = 40; rep = 2
nu = 40; rep = 3
nu = 40; rep = 4
nu = 40; rep = 5
nu = 40; rep = 6
nu = 40; rep = 7
nu = 40; rep = 8
nu = 40; rep = 9
nu = 50; rep = 0
nu = 50; rep = 1
nu = 50; rep = 2
nu = 50; rep = 3
nu = 50; rep = 4
nu = 50; rep = 5
nu = 50; rep = 6
nu = 50; rep = 7
nu = 50; rep = 8
nu = 50; rep = 9
nu = 60; rep = 0
nu = 60; rep = 1
nu = 60; rep = 2
nu = 60; rep = 3
nu = 60; rep = 4
nu = 60; rep = 5
nu = 60; rep = 6
nu = 60; rep = 7
nu = 60; rep = 8
nu = 60; rep = 9
nu = 70; rep = 0
nu = 70; rep = 1
nu = 70; rep = 2
nu = 70; rep = 3
nu = 70; rep = 4
nu = 70; rep = 5
nu = 70; rep = 6
nu = 70; rep = 7
nu = 70; rep = 8
nu = 70; rep = 9
nu = 80; rep = 0
nu = 80; rep = 1
nu = 80; rep = 2
nu = 80; rep = 3
nu = 80; rep = 4
nu = 80; rep = 5
nu = 80; rep = 6
nu = 80; rep = 7
nu = 80; rep = 8
nu = 80; rep = 9
nu = 90; rep = 0
nu = 90; rep = 1
nu = 90; rep = 2
nu = 90; rep = 3
nu = 90; rep = 4
nu = 90; rep = 5
nu = 90; rep = 6
nu = 90; rep = 7
nu = 90; rep = 8
nu = 90; rep = 9
nu = 100; rep = 0
nu = 100; rep = 1
nu = 100; rep = 2
nu = 100; rep = 3
nu = 100; rep = 4
nu = 100; rep = 5
nu = 100; rep = 6
nu = 100; rep = 7
nu = 100; rep = 8
nu = 100; rep = 9
nu = 120; rep = 0
nu = 120; rep = 1
nu = 120; rep = 2
nu = 120; rep = 3
nu = 120; rep = 4
nu = 120; rep = 5
nu = 120; rep = 6
nu = 120; rep = 7
nu = 120; rep = 8
nu = 120; rep = 9
nu = 150; rep = 0
nu = 150; rep = 1
nu = 150; rep = 2
nu = 150; rep = 3
nu = 150; rep = 4
nu = 150; rep = 5
nu = 150; rep = 6
nu = 150; rep = 7
nu = 150; rep = 8
nu = 150; rep = 9
nu = 200; rep = 0
nu = 200; rep = 1
nu = 200; rep = 2
nu = 200; rep = 3
nu = 200; rep = 4
nu = 200; rep = 5
nu = 200; rep = 6
nu = 200; rep = 7
nu = 200; rep = 8
nu = 200; rep = 9
nu = 500; rep = 0
nu = 500; rep = 1
nu = 500; rep = 2
nu = 500; rep = 3
nu = 500; rep = 4
nu = 500; rep = 5
nu = 500; rep = 6
nu = 500; rep = 7
nu = 500; rep = 8
nu = 500; rep = 9
nu = 1000; rep = 0
nu = 1000; rep = 1
nu = 1000; rep = 2
nu = 1000; rep = 3
nu = 1000; rep = 4
nu = 1000; rep = 5
nu = 1000; rep = 6
nu = 1000; rep = 7
nu = 1000; rep = 8
nu = 1000; rep = 9
nu = RCTD; rep = 0
nu = RCTD; rep = 1
nu = RCTD; rep = 2
nu = RCTD; rep = 3
nu = RCTD; rep = 4
nu = RCTD; rep = 5
nu = RCTD; rep = 6
nu = RCTD; rep = 7
nu = RCTD; rep = 8
nu = RCTD; rep = 9
```

--------------------------------

### Create spatial grid

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Generates a grid of points within the spatial data boundaries and visualizes it with scatter plots and dashed lines.

```python
x_min = sp_adata.obs['x'].min()
x_max = sp_adata.obs['x'].max()
y_min = sp_adata.obs['y'].min()
y_max = sp_adata.obs['y'].max()

num_x = 48
num_y = 48

spatial_adata =  sp_adata.copy()

# make grid 
x_seg = np.linspace(x_min, x_max, num_x)
y_seg = np.linspace(y_min, y_max, num_y)

with mpl.rc_context({'figure.figsize': (8, 8), 'figure.dpi': 150}):
    ax = sns.scatterplot(data=spatial_adata.obs, x="x", y="y", hue=sp_cell_class_column, s = 25, palette = color_dict, legend = False)
    ax.invert_yaxis()
    for i in range(x_seg.shape[0]):
        ax.plot([x_seg[i] for _ in y_seg],y_seg,linestyle='dashed', c = 'salmon', lw = 1, alpha = 0.5)
    for i in range(y_seg.shape[0]):    
        ax.plot(x_seg,[y_seg[i] for _ in x_seg],linestyle='dashed', c = 'salmon', lw = 1, alpha = 0.5)
#         ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))
    ax.axis('off')
```

--------------------------------

### Load AnnData Objects

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Loads two .h5ad files, slice1 and slice2, which have undergone nuclei segmentation.

```python
slice1 = sc.read(slice1_file)
slice2 = sc.read(slice2_file)
```

--------------------------------

### Import PASTE and load ST slices

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Imports the PASTE library for spatial alignment and defines the file paths for two Visium mouse brain cortex slices.

```python
import paste as pst
from scipy.spatial.distance import pdist
import matplotlib.patches as mpatches

slice1_file = '../demo_data/Visium_MouseBrain_Cortex_section1.h5ad'
slice2_file = '../demo_data/Visium_MouseBrain_Cortex_section2.h5ad'
```

--------------------------------

### Load scRef and sampled pseudo-cells

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Loads scRef data and sampled cells, sets the cell type key, and plots the data.

```python
ad_sc = sc.read('../Ckpts_scRefs/VISp/Ref_scRNA_VISp_qc2_2Kgenes.h5ad')
sampled_cells = sc.read('../Ckpts_scRefs/VISp/model_5000.h5ad') # 2K cells sampled from the learned gene expression distribution of scRef
cell_type_key='cell_subclass'
sns.set_context('paper',font_scale=1.5)
adata_all = PlotSampledData(sampled_cells,ad_sc,cell_type_key,palette=color_dict)
```

--------------------------------

### Benchmarking Loop Initialization

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Initializes lists and dataframes for storing error rates and deconvolution results during the benchmarking process.

```python
nu_list = [0.000001, 5] + np.arange(10,110,10).tolist() + [120, 150, 200, 500, 1000] + ['RCTD']

error_rate_df = pd.DataFrame(columns = {'error_rate', 'nu', 'rep'})
decon_df = pd.DataFrame()
decon_pcc_df = pd.DataFrame()
decon_rmse_df = pd.DataFrame()
```

--------------------------------

### Display first few rows of observation data

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Shows the first few rows of the observation metadata for the spatial dataset.

```python
sp_adata.obs.head()
```

--------------------------------

### Simulating Spots and Generating Data

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

Aggregates cells on uniform grids to simulate spots, generates fake spatial data with varying UMI counts, and filters spots with low UMI counts. It also prepares the data for saving.

```python
x_min = sp_adata.obs['X'].min()
x_max = sp_adata.obs['X'].max()
y_min = sp_adata.obs['Y'].min()
y_max = sp_adata.obs['Y'].max()

num_x = 16
num_y = 58

spatial_adata =  sp_adata.copy()

# make grid 
x_seg = np.linspace(x_min, x_max, num_x)
y_seg = np.linspace(y_min, y_max, num_y)

with mpl.rc_context({'figure.figsize': (3, 7)}):
    ax = sns.scatterplot(data=spatial_adata.obs, x="X", y="Y", hue=sp_cell_class_column, s = 25, palette = color_dict, legend = False)
    ax.invert_yaxis()
    for i in range(x_seg.shape[0]):
        ax.plot([x_seg[i] for _ in y_seg],y_seg,linestyle='dashed', c = 'salmon', lw = .1, alpha = 0.5)
    for i in range(y_seg.shape[0]):    
        ax.plot(x_seg,[y_seg[i] for _ in x_seg],linestyle='dashed', c = 'salmon', lw = .1, alpha = 0.5)
#         ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))
    ax.axis('off')
    
for sample_UMI in [130, 260, 520]:
    spatial_adata =  sp_adata.copy()
    
    # create fake spatial data with 'cell_locations' in uns (We assume the truth cell number in each fake spot is known in simulation study, thus skip "Nuclei segmentation" step in SpatialScope)
    spatial_adata = add_spot_label(spatial_adata, x_seg, y_seg)
    print(f'cell number range: {np.unique(generated_spot_adata.obs.cell_nums)}')
    generated_spot_adata = generate_spot_adata_func(spatial_adata, sample_UMI = sample_UMI)

    temp_obs = rename_cell_locations_obs(spatial_adata)
    generated_spot_adata.uns['cell_locations'] = temp_obs

    # filter spot that UMI counts smaller than 20
    generated_spot_adata_filtered = generated_spot_adata.copy()[~(generated_spot_adata.X.sum(1) < 20)]

    index = ~generated_spot_adata.uns['cell_locations']['spot_index_int'].isin(np.where(generated_spot_adata.X.sum(1) < 20)[0])
    generated_spot_adata_filtered.uns['cell_locations'] = generated_spot_adata.uns['cell_locations'][index].copy()


    # save data 
    save_data_dir = os.path.join(SpatialScope_base_path, f'demo_data/simulation/{int(sample_UMI)}UMI')
    ## spatial
    generated_spot_adata_filtered.obs = generated_spot_adata_filtered.obs.rename(columns = {'X':'x', 'Y':'y'})
    generated_spot_adata_filtered.obsm['spatial'] = generated_spot_adata_filtered.obs[['x','y']].values
    cell_locations = generated_spot_adata_filtered.uns['cell_locations'].copy()
    cell_locations = cell_locations.rename(columns = {'X':'x', 'Y':'y'})
    generated_spot_adata_filtered.uns['cell_locations'] = cell_locations
    generated_spot_adata_filtered.obs_names_make_unique()
    generated_spot_adata_filtered.var_names_make_unique()
    save_sp_dir = os.path.join(save_data_dir, 'spatial.h5ad')
    print(f'saved simulated spatial data of subsample {sample_UMI} UMI in {save_sp_dir}')
```

--------------------------------

### Data Preparation for Visualization

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Prepares a DataFrame for visualizing deconvoluted cell type proportions, combining results from different methods.

```python
temp = decon_df_temp.copy()
temp = temp[(temp.nu == 0)]
temp['nu'] = [str(0.000001)] * temp.shape[0]
temp['SpatialScope'] = small_nv_decon_df.SpatialScope.tolist()

temp = pd.concat([temp, decon_df_temp.copy()], axis = 0)
temp
```

--------------------------------

### StarDist + RCTD Comparison

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_1.ipynb

This code snippet shows how to run the StarDist + RCTD method for comparison, iterating through different repetitions.

```shell
for rep in {0..9}

do 

python ./compared_methods/SDRCTD.py \
--out_dir ./output/RCTDnv_MOp/260UMI/rep$rep \
--ST_Data ./demo_data/simulation/260UMI/spatial.h5ad \
--SC_Data ./Ckpts_scRefs/MERFISH_MOp_4Kcells/MERFISH_MOp_4Kcells.h5ad \
--cell_class_column subclass \
--cell_num_column cell_nums \
--hs_ST \

done

```

--------------------------------

### Initialization and Data Loading

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_6.ipynb

Initializes nu_list and loads decon_df, decon_pcc_df, and decon_rmse_df from CSV files. It then iterates through nu_list and repetitions.

```python
nu_list = [0]

decon_df = pd.read_csv(os.path.join(DataDir, 'decon_df.csv'), index_col = 0)
decon_pcc_df = pd.DataFrame()
decon_rmse_df = pd.DataFrame()

for j, nu in enumerate(nu_list):
    for i, rep in enumerate(range(10)):
        print(f"nu = {nu}; rep = {rep}")
        
        ################################################################################################################## RCTD resutls
        with open(os.path.join(DataDir, neighbor_range, 'rep' + str(rep), 'InitProp.pickle'), 'rb') as handle:
            InitProp = pickle.load(handle)

        RCTD_results = (InitProp['results'] / np.array(InitProp['results'].sum(1)))[:, None])
        RCTD_results = RCTD_results.loc[:, np.unique(RCTD_results.columns)]
        ################################################################################################################## StarDist + RCTD resutls
        SR_results = sc.read(os.path.join(DataDir, neighbor_range, 'rep' + str(rep), 'SDRCTD_results/single_cell_type_label_bySDRCTD.h5ad'))
        SR_cell_type_label = SR_results.uns['cell_locations'].copy()
        prop = label_to_matrix(np.array(SR_cell_type_label['spot_index_int']), np.array(SR_cell_type_label['SDRCTD_cell_type']))
        prop = prop/prop.sum(1)[:,None]

        SR_results = prop
        SR_results.index = np.arange(len(SR_results))
        SR_results = SR_results.loc[:,np.unique(SR_results.columns)]
    ################################################################################################################## SpatialScope_3D resuslts
        # palette = ['#F0EFF6', '#D0E2F8', '#c44e52']
        cell_locations = pd.read_csv(os.path.join(DataDir, neighbor_range, 'rep' + str(rep), 'CellTypeLabel_nu' + str(nu) + '.csv'), index_col = 0)

        prop = label_to_matrix(np.array(cell_locations['spot_index_int']), np.array(cell_locations['discrete_label_ct']))
        prop = prop/prop.sum(1)[:,None]

        SpatialScope_results = prop
        SpatialScope_results.index = np.arange(len(RCTD_results))
        SpatialScope_results = SpatialScope_results.loc[:,np.unique(SpatialScope_results.columns)]
        print(os.path.join(DataDir, neighbor_range, 'rep' + str(rep), 'SpatialScope_result'+ '_nu' + str(nu) +'.txt'))
        SpatialScope_results.to_csv(os.path.join(DataDir, neighbor_range, 'rep' + str(rep), 'SpatialScope_result'+ '_nu' + str(nu) +'.txt'))  


   ################################################################################################################## SpatialScope resuslts     
        # palette = ['#F0EFF6', '#D0E2F8', '#c44e52']
        cell_locations = pd.read_csv(os.path.join(DataDir, '3D_single', neighbor_range, 'rep' + str(rep), 'CellTypeLabel_nu' + str(nu) + '.csv'), index_col = 0)

        prop = label_to_matrix(np.array(cell_locations['spot_index_int']), np.array(cell_locations['discrete_label_ct']))
        prop = prop/prop.sum(1)[:,None]

        SpatialScope_3D_single_results = prop
        SpatialScope_3D_single_results.index = np.arange(len(SpatialScope_3D_single_results))
        SpatialScope_3D_single_results = SpatialScope_3D_single_results.loc[:,np.unique(SpatialScope_3D_single_results.columns)]

    ##################################################################################################################  
        RCTD_results = RCTD_results.loc[:,gd_results.columns]
        SR_results = SR_results.loc[:,gd_results.columns]
        SpatialScope_results = SpatialScope_results.loc[:,gd_results.columns]
        SpatialScope_3D_single_results = SpatialScope_3D_single_results.loc[:,gd_results.columns]  


        starmap_spots_pcc = compare_results(
            gd_results,
            [RCTD_results,SR_results, SpatialScope_3D_single_results, SpatialScope_results],
            columns = ['RCTD', 'SR', 'SpatialScope', 'SpatialScope_3D'],
            axis=0,
            metric='pcc'
        )

        starmap_spots_rmse = compare_results(
            gd_results,
            [RCTD_results,SR_results, SpatialScope_3D_single_results, SpatialScope_results],
            columns = ['RCTD', 'SR', 'SpatialScope', 'SpatialScope_3D'],
            axis=0,
            metric='rmse'
        )

        starmap_spots_pcc['nu'] = [int(nu)] * starmap_spots_pcc.shape[0]
        starmap_spots_pcc['metric'] = ['pcc'] * starmap_spots_pcc.shape[0]
        starmap_spots_rmse['nu'] = [int(nu)] * starmap_spots_rmse.shape[0]
        starmap_spots_rmse['metric'] = ['rmse'] * starmap_spots_rmse.shape[0]

        starmap_spots_pcc['cell_nums'] = starmap_spots_rmse['cell_nums'] = sp_adata.obs.cell_nums.tolist()

        decon_df = pd.concat([decon_df, starmap_spots_pcc, starmap_spots_rmse], axis = 0)
        decon_pcc_df = pd.concat([decon_pcc_df, starmap_spots_pcc], axis = 0)
        decon_rmse_df = pd.concat([decon_rmse_df, starmap_spots_rmse], axis = 0)


        

```

--------------------------------

### Load Spatial Datasets

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Mouse-Brain.ipynb

Loads two sections of Visium Mouse Brain Cortex data using scanpy.

```python
slice1_file = '../demo_data/Visium_MouseBrain_Cortex_section1.h5ad'
slice2_file = '../demo_data/Visium_MouseBrain_Cortex_section2.h5ad'
slice1 = sc.read(slice1_file)
slice2 = sc.read(slice2_file)
```

--------------------------------

### Displaying DataFrame

Source: https://github.com/yanglabhkust/spatialscope/blob/master/demos/Benchmarking-Dataset_2.ipynb

Displays the content of the 'test_df' DataFrame.

```python
test_df
```
Tokens:
Raw