networkvi 0.1.0

Biologically Guided Variational Inference for Interpretable Multimodal Single-Cell Integration and Mechanistic Discovery

NetworkVI: Biologically Guided Variational Inference for Interpretable Multimodal Single-Cell Integration and Mechanistic Discovery

Getting started

NetworkVI is a sparse deep generative model designed for the paired, vertical (shared cells across measurements), horizontal (shared features across datasets) or mosaic integration and interpretation of multimodal single-cell data. The model learns a rich, batch-corrected low-dimensional representation of bi- and trimodal single-cell count datasets, estimating the representation using normalized input data. Please refer to the documentation. We also provide tutorials:

• Paired integration and query-to-reference mapping
• Mosaic integration
• Interpretability: Inference of GO importances and Gene-GO associations
• Interpretability: Infernce of GO term-specific covariate attention values

Installation

NetworkVI requires Python>3.9 on your system.

1. Install the latest release of NetworkVI from PyPi:

pip install networkvi

2. Install the latest development version:

pip install git+https://github.com/LArnoldt/networkvi.git@main

Please also install the appropiate CUDA version of torch, torch-scatter and torch-sparse version. Here we give an example for CUDA 12.1:

pip install -U torch==2.2.0 --index-url https://download.pytorch.org/whl/cu121
pip install -U torch-scatter torch-sparse -f https://data.pyg.org/whl/torch-2.2.0+cu121.html

API

Please find the API here.

Release notes

Please find the release notes here.

Contact

If you found a bug, please use the issue tracker. If you use NetworkVI in your research, please consider citing the preprint:

Arnoldt, L., Upmeier zu Belzen, J., Herrmann, L., Nguyen, K., Theis, F.J., Wild, B. , Eils, R., "Biologically Guided Variational Inference for Interpretable Multimodal Single-Cell Integration and Mechanistic Discovery", bioRxiv, June 2025.

Reproducibility

Code and notebooks to reproduce the results and figues from the paper are available here.