Installation#

Prerequisites#

scvi-tools can be installed via conda or pip. If you don’t know which to choose, we recommend conda.

conda#

  1. Install conda. We typically use the mambaforge distribution. Use python>=3.9, and consider using mamba instead of conda. Mamba is a drop-in replacement for conda that is significantly more efficient.

  2. Create a new conda environment:

    conda create -n scvi-env python=3.9
    
  3. Activate your environment:

    conda activate scvi-env
    

pip#

  1. If using conda/mamba, then just run conda install -c anaconda pip and skip this section.

  2. Install Python, we prefer the pyenv version management system, along with pyenv-virtualenv.

  3. Install PyTorch and jax. If you have an Nvidia GPU, be sure to install versions of PyTorch and jax that support it – scvi-tools runs much faster with a discrete GPU.

Apple silicon#

Installing scvi-tools on a Mac with Apple Silicon is only possible using a native version of python. A native version of python can be installed with an Apple Silicon version of mambaforge (which can be installed from a native version of homebrew via brew install --cask mambaforge). This is due to an scvi-tools dependency on jax, which cannot be run via Rosetta.

Windows#

After setting up a virtual environment with conda/mamba/pyenv, please install the community-built version of jax before installing scvi-tools.

pip install "jax[cpu]" -f https://whls.blob.core.windows.net/unstable/index.html --use-deprecated legacy-resolver

GPU#

All scvi-tools models will be faster when accelerated with a GPU. Before installing scvi-tools, you can install GPU versions of PyTorch and jax using conda as follows:

conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia
conda install jax jaxlib -c conda-forge

Please go to the respective package website for more information on how to install with pip.

Conda#

conda install scvi-tools -c conda-forge

Pip#

pip install scvi-tools

Through pip with packages to run notebooks. This installs scanpy, etc.:

pip install "scvi-tools[tutorials]"

Nightly version - clone this repo and run:

pip install .

Development#

For development - clone this repo and run:

pip install -e ".[dev,docs]"

R#

scvi-tools can be called from R via Reticulate.

This is only recommended for basic functionality (getting the latent space, normalized expression, differential expression). For more involved analyses with scvi-tools, we highly recommend using it from Python.

The easiest way to install scvi-tools for R is via conda.

  1. Install conda prerequisites.

  2. Install R and reticulate in the conda environment:

    conda install -c conda-forge r-base r-essentials r-reticulate
    
  3. Then in your R code:

    library(reticulate)