dgscrna 1.1

A Python package for single-cell RNA-seq cell type annotation using marker-based scoring and deep learning

DGscRNA

A Python package for single-cell RNA-seq cell type annotation using marker-based scoring and deep learning refinement.

Overview

DGscRNA combines traditional marker-based cell type scoring with deep learning to resolve ambiguous cell type assignments in single-cell RNA-seq data. The workflow includes:

1. Preprocessing: Quality control, normalization, and dimensionality reduction
2. Clustering: Multiple clustering algorithms (Leiden, HDBSCAN, K-means)
3. Marker Scoring: Density-based scoring using known cell type markers
4. Deep Learning: Neural network refinement of ambiguous annotations

Installation

pip install dgscrna

Or install from source:

git clone https://github.com/yourusername/DGscRNA.git
cd DGscRNA
pip install -e .

Quick Start

import scanpy as sc
import dgscrna as dg

# Load your data
adata = sc.read_h5ad('your_data.h5ad')

# Run the complete pipeline
results = dg.run_dgscrna_pipeline(
    adata=adata,
    marker_folder='path/to/marker/sets/',
    clustering_methods=['leiden', 'hdbscan'],
    deep_learning=True
)

# View results
sc.pl.umap(adata, color=['leiden', 'CellMarker_Thyroid_mean_DGscRNA'])

Input Data Format

Single-cell Data

• Format: AnnData object (scanpy/anndata)
• Requirements: Preprocessed and normalized gene expression matrix

Marker Sets

• Format: CSV files in a folder
• Structure: Columns are cell type names, rows are marker genes
• Example:

,CellType1,CellType2,CellType3
0,Gene1,Gene4,Gene7
1,Gene2,Gene5,Gene8
2,Gene3,Gene6,Gene9

Output

• AnnData object: With added annotation columns
• Results dictionary: Training scores and metrics
• Visualization: UMAP plots with annotations

Documentation

• API Reference
• Installation Guide
• Tutorial
• Examples

License

GPL-3.0 License - see LICENSE file for details.

Contributing

Contributions are welcome! Please read our contributing guidelines and submit pull requests.

Support

For questions and support, please open an issue on GitHub or contact the maintainers.