cellpath 0.1.dev0

CellPath, multiple trajectories inference in single cell RNA-Seq data from RNA velocity

CellPath(Inference of multiple trajectories in single cell RNA-Seq data from RNA velocity)

CellPath v0.1.0

Zhang's Lab, Georgia Institute of Technology

Developed by Ziqi Zhang

Description

CellPath is a single cell trajectory inference method that infers cell developmental trajectory using single-cell RNA Sequencing data and RNA-velocity data. The preprint is posted on bioarxiv: https://www.biorxiv.org/content/10.1101/2020.09.30.321125v2

News

Include leiden algorithm for meta-cell clustering, which is more suitable for datasets with intricate trajectories. You can specify the clustering algorithm you use with either flavor = "leiden" or flavor = "k-means" in cellpath.meta_cell_construction() or cellpath.all_in_one(), please check the run_cellpath.ipynb for more details.

Dependencies

Python >= 3.6.0

numpy >= 1.18.2

scipy >= 1.4.1

pandas >= 1.0.3

scikit-learn >= 0.22.1

anndata >= 0.7.1

scvelo >= 0.2.0

seaborn >= 0.10.0

statsmodels >= 0.11.1 (optional, for differentially expressed gene analysis)

rpy2 >= 3.3.0 (optional, for principal curve only)

Installation

Clone the repository with

git clone https://github.com/PeterZZQ/CellPaths.git

And run

cd CellPaths/
pip install .

Uninstall using

pip uninstall cellpath

Usage

run_cellpath.ipynb provide a short pipeline of running cellpaths using cycle-tree trajectory dataset in the paper.

• Initialize using adata with calculated velocity using scvelo

cellpath_obj = cp.CellPath(adata = adata, preprocess = True)

preprocessing: the velocity has been calculated and stored in adata or not, if False, the velocity will be calculated during initialization with scvelo

• Run cellpath all in one

cellpath_obj.all_in_one(num_metacells = num_metacells, n_neighs = 10, pruning = False, num_trajs = num_trajs, insertion = True, prop_insert = 0.50)

num_metacells: number of meta-cells in total

n_neighs: number of neighbors for each meta-cell

pruning: way to construct symmetric k-nn graph, prunning knn edges or including more edges

num_trajs: number of trajectories to output in the end

insertion: insert unassigned cells to trajectories or not

prop_insert: proportion of cells to be incorporated into the trajectories

`Pseudo-time and branching assignment result

cellpath_obj.pseudo_order

• Additional visualizations, please check run_cellpath.ipynb for details.

Datasets

• You can access the real dataset that we used for the benchmarking through: https://www.dropbox.com/sh/nix4wnoiwda5id5/AACTxvGTQ82UzwMJs2IWSriKa?dl=0. You can reproduce the result by putting the file into the root directory and run the notebook in ./Examples/.

  • ./Examples/CellPath_hema.ipynb: mouse hematopoiesis dataset.
  • ./Examples/CellPath_dg.ipynb: dentate-gyrus dataset.
  • ./Examples/CellPath_pe.ipynb: pancreatic endocrinogenesis dataset.
  • ./Examples/CellPath_forebrain.ipynb: forebrain dataset.

Contents

• CellPath/ contains the python code for the package
• real_data/ contains three real datasets, used in the paper, dentate-gyrus dataset, pancreatic endocrinogenesis dataset and human forebrain dataset. Files in real_data folder can be downloaded from dropbox
• sim_data/ contains four simulated datasets, used in the paper, with trifurcating, double-batches, cycle-tree and multiple-cycles structure.
  • Trifurcating and double-batches datasets are generated using dyngen
  • cycle-tree and multiple-cycles datasets are generated using symsim.