BAVARIA is python package that implements a Batch-adversarial Variational auto-encoder with Negative Multinomial reconstruction loss for single-cell ATAC-seq analysis.
In particular, the model can be used to extract a latent feature representation of a cell which can be used for downstream analysis tasks, including cell cluster, cell identification, etc. The package is freely available under a GNU Lesser General Public License v3 or later (LGPLv3+)
You can install the package version v0.1.0 via
pip install https://github.com/BIMSBbioinfo/bavaria/archive/v0.1.0.zip
Alternatively, you can install the latest version from the master branch using
pip install git+https://github.com/BIMSBbioinfo/bavaria.git
BAVARIA offers a command line interface that fits an ensemble of BAVARIA models given a raw count matrix (-data) Subsequently, the model parameters and latent features are stored in the output directory (-output)
bavaria -data adata.h5ad \
-output <outputdir> \
-epochs 200 \
-nrepeats 10 \
-nlatent 15 \
-batchnames batch \
-modelname bavaria
Additional information on available hyper-parameters are available through
bavaria -h
Below you find links to the tutorials.
The tutorials will require jupyter and other resources which are defined in
tutorial/requirements.txt. Using the requirements file you instantiate
a new conda environment using
conda create --name bavaria_tutorial --file tutorial/requirements.txt| Example notebooks |
|---|
| Data preparation PBMC integration |
| Using BAVARIA to integrate PBMC data |