scVI

Single-cell Variational Inference

• Free software: MIT license
• Documentation: https://scvi.readthedocs.io.

Quick Start

1. Install Python 3.6 or later. We typically use the Miniconda Python distribution.

2. Install PyTorch. If you have an Nvidia GPU, be sure to install a version of PyTorch that supports it -- scVI runs much faster with a discrete GPU.

3. Install scvi through conda (conda install scvi -c bioconda) or through pip (pip install scvi). Alternatively, you may clone this repository and manually install the dependencies listed in setup.py.

4. Refer to this Jupyter notebook to see how to import datasets into scVI.

5. Refer to this Jupyter notebook to see how to train the scVI model, impute missing data, detect differential expression, and more!

Benchmarks

To recreate the results appearing in the paper referenced below, run

python ./run_benchmarks.py --dataset=cortex

Valid choices for --dataset include synthetic, cortex, brain_large, retina, cbmc, hemato, and pbmc. You may also specify an arbitrary .loom, .h5ad (AnnData), or .csv file.

References

Romain Lopez, Jeffrey Regier, Michael B Cole, Michael Jordan, Nir Yosef. "Bayesian Inference for a Generative Model of Transcriptome Profiles from Single-cell RNA Sequencing." In submission. Preprint available at https://www.biorxiv.org/content/early/2018/03/30/292037