Installation and Setup#

Installing SIGnature#

The SIGnature API is under activate development. The latest development API can be downloaded from GitHub and installed as follows:

git clone https://github.com/genentech/signature.git
cd signature
pip install -e .

Downloading precalculated attributions#

Download the cell type-level attributions from Zenodo: https://zenodo.org/communities/signature/

Downloading models#

To generate attribution scores on new data, helper files can be downloaded from Zenodo: https://zenodo.org/records/17903196

SIGnature currently supports calculating attributions using the following models:

  1. SCimilarity: pretrained weights included in helper files, but can also be downloaded here: https://zenodo.org/records/15729925

  2. scFoundation: pretrained weights can be downloaded here: https://huggingface.co/genbio-ai/scFoundation/tree/main

  3. scVI: pretrained weights for scVI models trained on CZI Census data can be downloaded here: https://cellxgene.cziscience.com/census-models

  4. SSL-scTab: pretrained weights for self-supervised learning models trained on scTab data can be downloaded here: https://huggingface.co/TillR/sc_pretrained/tree/main

Conda environment setup#

To install SIGnature in a [Conda](https://docs.conda.io) environment we recommend this environment setup:

Download environment file

name: signature
channels:
  - nvidia
  - pytorch
  - bioconda
  - conda-forge
dependencies:
  - python=3.10
  - ipykernel
  - ipython
  - ipywidgets
  - pip

Followed by installing the signature package, as above.