scPlantLLM

scPlantLLM(Single-cell Plant Large Language Model), a transformer-based model specifically designed for the exploration of single-cell expression atlases in plants

This project aims to pretrain a large model using 1 million plant single-cell data and further pretrain a cell annotation model and a batch integration model based on this pretrained model. Ultimately, we can use these pretrained models for cell annotation, providing an efficient method for handling large-scale plant single-cell data.

Stage 1: Data Extraction

Requirements

• R ≥4.0
• Seurat package
• Python 3.11+
• Refer to requirements.txt for the list of dependencies.

Extract Data from RDS Files

Run the following command to extract data from RDS files and generate .h5 and meta.csv files:

Rscript extract_rds_data.R ./data/raw ./data/processed

• ./data/raw: Path to the input RDS data.
• ./data/processed: Path to the output .h5 and meta.csv files.

If fine-tuning the model is required, meta.csv can be used to construct the celltype.meta file.

Stage 2: Generate Metadata Information

Handling Batch Effect

Run the script:

python prepare_meta.py \
    --input_path ./data/processed \
    --output_path ./data/processed/has_celltype \
    --file_prefix batch_effect \
    --col_name orig.ident \
    --do_batch

Output:

├── batch_effect.meta
├── batch_effect_vocab.meta.json

Handling Cell Type Information

Run the script:

python prepare_meta.py \
    --input_path ./data/processed \
    --output_path ./data/processed/has_celltype \
    --file_prefix cell_type \
    --col_name seurat_clusters \
    --do_cell_type

Output:

├── cell_type.meta
├── cell_type_vocab.meta.json

Stage 3: Build Model Input Data

Without Cell Type Information

python preprocess_data.py \
    --input_path ./data/processed \
    --output_path ./data/processed/dont_have_celltype \
    --gene_vocab_file ./config/gene_vocab.json \
    --batch_effect_file ./data/processed/has_celltype/batch_effect.meta \
    --batch_effect_vocab_file ./data/processed/has_celltype/batch_effect_vocab.meta.json

With Cell Type Information

python preprocess_data.py \
    --input_path ./data/processed \
    --output_path ./data/processed/has_celltype \
    --gene_vocab_file ./config/gene_vocab.json \
    --has_celltype \
    --cell_type_file ./config/celltype_meta.csv \
    --cell_type_vocab_file ./config/celltype_vocab.json \
    --batch_effect_file ./data/processed/has_celltype/batch_effect.meta \
    --batch_effect_vocab_file ./data/processed/has_celltype/batch_effect_vocab.meta.json \
    --test_size 0.1

Stage 4: Use the Model

The model usage is detailed in the tutorial notebook:

Tutorial.ipynb

Citation

@article{10.1093/gpbjnl/qzaf024,
    author = {Cao, Guangshuo and Chao, Haoyu and Zheng, Wenqi and Lan, Yangming and Lu, Kaiyan and Wang, Yueyi and Chen, Ming and Zhang, He and Chen, Dijun},
    title = {Harnessing the Foundation Model for Exploration of Single-cell Expression Atlases in Plants},
    journal = {Genomics, Proteomics & Bioinformatics},
    pages = {qzaf024},
    year = {2025},
    month = {03},
    issn = {1672-0229},
    doi = {10.1093/gpbjnl/qzaf024},
    url = {https://doi.org/10.1093/gpbjnl/qzaf024},
}