scPlantFormer

This project provides tools for cross-dataset cell-type annotation, pre-training models on multi-modal datasets (scRNA-seq, scATAC-seq), and integrating datasets across species, tissues, and experimental conditions. The repository includes a set of Python scripts and Jupyter notebooks designed to facilitate the integration and analysis of large-scale biological data.

Features

• Pre-training models on large-scale multi-modal datasets for improved cell-type prediction and annotation.
• Fine-grained cell-type annotation using custom attention-based deep learning models implemented in PyTorch.
• Jupyter notebooks for integration of single-cell data across different batches, species, tissues, and modalities (e.g., scRNA-seq and scATAC-seq).

Project Structure

• Pretrained_Model/: Directory for storing models.

  • Arabidopsis_all_Pretrained.pth: Pre-trained model on All Arabidopsis scRNA-seq data.
  • Flower_Pretrained.pth: Pre-trained model on All Arabidopsis'flower scRNA-seq data.
  • Leaf_Pretrained.pth: Pre-trained model on All Arabidopsis'leaf scRNA-seq data.
  • Root_Pretrained.pth: Pre-trained model on All Arabidopsis'root scRNA-seq data.
  • seed_Pretrained.pth: Pre-trained model on All Arabidopsis'seed scRNA-seq data.

• Tutorial/: Jupyter notebooks for various data integration tasks:

  • Integration_batch.ipynb: Integration of datasets across different experimental batches.
  • Integration_scRNA_scATAC.ipynb: Integration of single-cell RNA-seq (scRNA-seq) and single-cell ATAC-seq (scATAC-seq) data.
  • Integration_species.ipynb: Integration of single-cell data across different species.
  • Integration_tissues.ipynb: Integration of single-cell data across different tissues.
  • cross_dataset_cell-type_annotation.py: Script for cross-dataset cell-type annotation using machine learning models (e.g., Logistic Regression, Decision Trees, VotingClassifier, StackingClassifier). This script processes large-scale biological data, performs classification, and logs the results.
  • inner_cell_type_annotation.py: Implements deep learning models for inner cell-type annotation using attention-based layers (e.g., CausalSelfAttention). The script leverages PyTorch and is designed for neural network-based analysis of single-cell data, including configuration management and random seed setting.

• model/: Python scripts for cell-type annotation and model training:

  • pre_train_all.py: Script for pre-training models across different single-cell datasets, using custom neural network layers for embedding and processing the data. It is optimized for large-scale data, including plant-specific models such as scPlantGPT_v1.

Installation

To set up the project and install the required packages:

1. Clone the repository:

   git clone https://github.com/your_username/project_name.git
   cd project_name
   

requirements.txt

torch
scanpy
scikit-learn
geomloss
tqdm
numpy
pandas