Skip to content

Preprocess

1. Data aquisition

We categorize the preprocessing of plant ST data into four parts: data acquisition, defining spatial units, constructing a spatial matrix with coordinates, and quality control.

Here we use the data from the paper "Spatiotemporal transcriptomic landscape of rice embryonic cells during seed germination", it is a rice seed embryo spatial transcriptomics dataset from stereo-seq. - The data has been aligned, quantified and barcode mapped through the SAW workflow - The data input here is GEM file, which is the output of the SAW workflow. - The dataset is available at Spatial Transcript Omics DataBase, STOmics DB STT0000049 - The pipiline stereppy is used to process the data. It accepts the GEM/GEF file (from Sterep-seq) and H5ad (output from scanpy) as input.

import sys
import os
import warnings
import stereo as st
warnings.filterwarnings('ignore')

data_path = '/home/yaoj/Review_plant_Spatial_multiomics_data_analysis/data/A5_1_bin1.gem.gz'
# read GEM file information
st.io.read_gem(data_path)

2. Define spatial units

2.1 Bin segmentation

Bin segmentation is simpler and more widely adopted, where the key step involves determining the appropriate bin size. The average cell size within the tissue is estimated using image information, and tools like ImageJ, and CaseViewer are often employed. The bin size can then be optimized based on clustering results, ensuring the clusters correspond to distinct cell types and are biologically interpretable

We estimate the average cell size in this rice seed embryo dataset is 25 μm². - Cell side length = X * ST_resolution - X is the bin size, in Bin 50, stereo-seq resolution is 0.5 um, the cell side length is 50 * 0.5 = 25 um

# define the bin size= 50
data = st.io.read_gem(file_path=data_path, bin_size=50)
data
StereoExpData object with n_cells X n_genes = 2566 X 25320 bin_type: bins bin_size: 50 offset_x = 10025 offset_y = 8025 cells: ['cell_name'] genes: ['gene_name'] cells_matrix = ['spatial'] Layers with keys: result: []

#data.tl.cal_qc()
data.plt.spatial_scatter()

[2025-01-18 03:31:51][Stereo][433552][MainThread][140065997267328][st_pipeline][41][INFO]: start to run cal_qc...
[2025-01-18 03:31:52][Stereo][433552][MainThread][140065997267328][st_pipeline][44][INFO]: cal_qc end, consume time 0.1649s.

BokehModel(combine_events=True, render_bundle={'docs_json': {'eee6ab55-1588-4559-888b-5eee0a19dc22': {'defs': …

png

Test bin size = 30,50,80,100, and compare the results

import panel as pn
def compare_multiple_bins(file_path, bin_sizes=[30, 50, 80, 100]):
    plots = []  # 存储所有图形对象
    for bin_size in bin_sizes:
        # 读取数据
        data = st.io.read_gem(file_path=file_path, bin_size=bin_size)
        data.tl.cal_qc()

        # plot spatial scatter
        plot = data.plt.spatial_scatter(
            cells_key=['n_genes_by_counts'],
            width=2,
            height=2,
            title=[f'bin size = {bin_size}']
        )
        plots.append(plot)

    # use panel's Row layout to display all plots side by side
    return pn.Row(*plots)

# use the function
pn.extension()  # initialize panel
combined_plot = compare_multiple_bins(data_path)
combined_plot

Bin size

2.2 Cell segmentation

Cell segmentation offers a more granular approach, especially in tissues where cell size exhibit considerable variability. Recent advancements in segmentation tools allow for more precise delineation of individual cells, typically using cell wall staining images and manually annotated training models. Popular tools include Cellpose2/3, StarDist. Comprehensive ST frameworks, such as STCellBin, support segmentation based on models trained with Arabidopsis seed cell segmentation ground truth.

##cell segmentation
#cellpose installation
#download environment.yml from https://github.com/MouseLand/cellpose
#cmd
conda env create -f environment.yml
conda activate cellpose
pip install cellpose


#prediction:
python -m cellpose  --exclude_on_edges  --dir .\data4test\img --diameter 0  --save_outlines  --save_txt  --in_folders --verbose --flow_threshold 5 --pretrained_model .\model\cellpose_residual_on_style_on_concatenation_off_CP_1_2022_11_26_02_48_20.469512

#--dir  (only slice images saving path)
#--save_txt   ("outlices.txt" whice serves as cell segmentation result in next step ,will be saved in a folder which is named "txt_outlines")
#--pretrained_model  (path of auto-segmentation model)
Cellsegmentation

3. Construct a spatial matrix with coordinates

4. Quality control and data optimization

4.1 Quality control

4.2 data imputation

4.3 batch correction