Description

This package provides a scalable Python suite for fast tree inference and advanced pseudotime downstream analysis, with a focus on fate biasing. This package is compatible with anndata object format used in scanpy or scvelo pipelines. A complete documentation of this package is available here.

Tree inference algorithms

The user have the choice between two algorithm for tree inference:

ElPiGraph

For scFates, the python implementation of the ElPiGraph algorithm is used, which include GPU accelerated principal tree inference. A self-contained description of the algorithm is available here or in the related paper

A R implementation of this algorithm is also available, coded by Luca Albergante

A native MATLAB implementation of the algorithm (coded by Andrei Zinovyev and Evgeny Mirkes) is also available

Simple PPT

A simple PPT inspired approach, translated from the crestree R package, code has been also adapted to run on GPU for accelerated tree inference.

Citations

Code for PPT inference and most of downstream pseudotime analysis was initially written in a R package by Ruslan Soldatov for the following paper:

Soldatov, R., Kaucka, M., Kastriti, M. E., Petersen, J., Chontorotzea, T., Englmaier, L., … Adameyko, I. (2019).
Spatiotemporal structure of cell fate decisions in murine neural crest.
Science, 364(6444).

if you are using ElPiGraph, please cite:

Albergante, L., Mirkes, E. M., Chen, H., Martin, A., Faure, L., Barillot, E., … Zinovyev, A. (2020).
Robust And Scalable Learning Of Complex Dataset Topologies Via Elpigraph.
Entropy, 22(3), 296.

Code for preprocessing has been translated from R package pagoda2, if you use any of these functions (scf.pp.batch_correct & scf.pp.find_overdispersed), please cite:

Nikolas Barkas, Viktor Petukhov, Peter Kharchenko and Evan
Biederstedt (2021). pagoda2: Single Cell Analysis and Differential
Expression. R package version 1.0.2.

Palantir python tool provides a great dimensionality reduction method, which usually lead to consitent trees with scFates, if use scf.pp.diffusion, please cite:

Manu Setty and Vaidotas Kiseliovas and Jacob Levine and Adam Gayoso and Linas Mazutis and Dana Pe'er (2019)
Characterization of cell fate probabilities in single-cell data with Palantir.
Nature Biotechnology

Installation

scFates is available on pypi, you can install it using:

pip install -U scFates

or the latest development version can be installed from GitHub:

pip install git+https://github.com/LouisFaure/scFates

Python dependencies

scFates gives the choice of between SimplePPT and ElPiGraph for learning a principal graph from the data. Elpigraph needs to be installed from its github repository with the following command:

pip install git+https://github.com/j-bac/elpigraph-python.git

R dependencies

scFates rely on the R package mgcv to perform testing and fitting of the features on the peudotime tree. Package is installed in an R session with the following command:

install.packages('mgcv')

GPU dependencies (optional)

If you have a nvidia GPU, scFates can leverage CUDA computations for speedups in some functions, for that you will need Rapids installed. This can be easily done via a dedicated conda environment:

conda create -n scFates-gpu -c rapidsai -c nvidia -c conda-forge -c defaults \
    rapids=0.19 python=3.8 cudatoolkit=11.0 -y
conda activate scFates-gpu
pip install git+https://github.com/j-bac/elpigraph-python.git
pip install scFates --ignore-installed

Docker container

scFates can be run on a Docker container based on Rapids 0.18 container, which provide a gpu enabled environment with Jupyter Lab. Use the following command:

docker run --rm -it --gpus all -p 8888:8888 -p 8787:8787 -p 8786:8786 \
    louisfaure/scfates:version-{version.number}