- Mubind is a machine learning method for learning motif associations with single cell genomics data, using graph representations such as a k-nearest neighbors graph (kNN).
- It uses sample-sample operation (graphs) to communicate filter activities (learned motifs) across cells.
- The whole codebase is written in PyTorch.
- This package works with single-cell genomics data, scATAC-seq, scChIP-seq, etc. We have also tested it on bulk in vitro samples (HT-SELEX, PBM). Please see the documentation for related examples.
- Number of cells: The scalability of this method has been tested on single-cell datasets between 10,000 and 100,000 cells.
- Number of peaks: We have tested randomly selected features, or EpiScanpy's variability score. Modeling all features requires calibration of batch sizes and total GPU memory.
- Usual running times (one GPU): We get variable running times based on hyper-parameters, and they range between 10 minutes (prior filters) and 3 hours (de novo).
Please refer to the documentation.
There are several alternative options to install mubind:
- Install the latest release of
mubind
fromPyPI <https://pypi.org/project/mubind/>
_:
pip install mubind
- Install the latest development version:
pip install git+https://github.com/theislab/mubind.git@main
Available soon.
See the changelog.
If you found a bug, please open an Issue.
If mubind is useful for your research, please consider citing as:
@software{mubind,
author = {Ibarra, Schneeberger, Erdogan, Martens, Aliee, Klein and Theis FJ},
doi = {},
month = {},
title = {{mubind}},
url = {https://github.com/theislab/mubind},
year = {2023}
}
t.b.c.
Project template created using scverse cookie template