Single-cell-resolution spatial transcriptomics for plant tissue: cell-wall segmentation → cell-level count matrix → Seurat analysis.
Scripts and a trained segmentation model used in
Yao J†, Chu Q†, Guo X†, Shao W†, et al., Liu H*, Fan L*. Spatiotemporal transcriptomic landscape of rice embryonic cells during seed germination. Developmental Cell (2024). doi:10.1016/j.devcel.2024.05.016
Plant cells have walls and vary widely in size, so bin-based spatial transcriptomics (e.g. Stereo-seq bin50) does not map onto real cells. This pipeline segments individual cells from a stained tissue image with a custom Cellpose model trained on plant cell walls, assigns every spatial barcode (DNB) to the cell it falls in, and produces a standard 10x-style count matrix that can be analysed as single-cell data.
面向植物组织的单细胞分辨率空间转录组流程：用针对植物细胞壁训练的 Cellpose 模型分割单个细胞，将 Stereo-seq 的空间条形码归属到细胞，生成可直接用 Seurat 分析的细胞级表达矩阵。配套 Developmental Cell 2024 水稻胚萌发时空转录组论文。
| Step | File | Input → Output |
|---|---|---|
| 1. Cell segmentation | 01cellposePrediction.txt |
tissue image (data4test/img/) → cell outlines (txt_outlines/*_cp_outlines.txt) using model/cellpose_* |
| 2. Barcodes → cells | 02outline2scResGemCsv.py |
outlines + image + Stereo-seq .gem → *_scResGem.csv (each DNB labelled with its cell ID) |
| 3. Build matrix | 03scResGemCsv2mtx.sh |
*_scResGem.csv → MatrixMarket folder (matrix4test/: matrix.mtx, genes.tsv, barcodes.tsv) + cell-centroid table |
| 4. Analysis | 04scST_SCT4test.R |
matrix + centroids → QC (mt/pt genes), SCTransform, clustering, spatial plots (scSTplot/) |
colorTools.py is a small RGB⇄int helper used by step 2. data4test/ holds a down-sized rice-embryo example so every step can be run end-to-end.
# 1. Segment cells (Cellpose ≥ 2.0; see 01cellposePrediction.txt for the conda env)
python -m cellpose --dir ./data4test/img --pretrained_model ./model/cellpose_residual_on_style_on_concatenation_off_CP_1_2022_11_26_02_48_20.469512 \
--diameter 0 --flow_threshold 5 --exclude_on_edges --save_outlines --save_txt --in_folders --verbose
# 2. Assign DNBs to cells (edit the paths at the top of the script)
python 02outline2scResGemCsv.py
# 3. Write a 10x-style matrix
bash 03scResGemCsv2mtx.sh
# 4. Analyse in R (Seurat ≥ 4)
Rscript 04scST_SCT4test.RDependencies. Python 3.8+: cellpose, opencv-python, pillow, numpy, pandas, matplotlib, tqdm. R ≥ 4.1: Seurat, SeuratObject, dplyr, ggplot2, ggpubr, ggthemes, Matrix.
model/cellpose_residual_on_style_on_concatenation_off_CP_1_2022_11_26_02_48_20.469512 is a Cellpose model fine-tuned on annotated rice embryo cell-wall images. It may transfer to other plant tissues with visible cell walls; if your staining or tissue differs, fine-tune it further on a few annotated slices.
If you use this code or model, please cite the Developmental Cell paper above.
MIT © 2022–2026 Jie Yao