EpiPack

Description

EpiPack is a modular deep learning toolkit for single-cell ATAC-seq reference mapping, cell label annotation, and out-of-reference (OOR) detection.
By introducing heterogeneous transfer learning and peak-informed variational inference (PEIVI), EpiPack enables scalable construction of harmonized reference atlases and robust query mapping across diverse scATAC-seq datasets. It further provides global-local OOR detection frameworks for discovering novel cell types or perturbed cellular states with interpretable uncertainty estimation. Please see our manuscript for more details.

Installation

The package is available on PyPI and can be installed with all required dependencies via:

pip install epipackpy

Tutorial

Please refer to our full documentation and tutorials at
👉 epipack.readthedocs.io

Demo datasets used in the tutorial can be downloaded from our Google Drive folder. The names of the datasets are aligned with the names that we used in the tutorial for better reproduction.

Dependencies

- Python >= 3.9  
- PyTorch >= 2.0.1  
- PyTorch-CUDA >= 11.8  
- NumPy >= 1.26.4  
- Pandas >= 1.5.3  
- SciPy >= 1.10.0  
- Scikit-learn >= 1.5.2  
- tqdm >= 4.66.1  
- Matplotlib >= 3.9.4  
- Seaborn >= 0.12.2  

For PyTorch installation, we recommend users to follow the official PyTorch installation guide to select the correct build based on their CUDA version.