Temporal RNA-seq Deconvolution (TeRNAdecov)

TeRNAdecov is a package for deconvolution of bulk RNA-seq samples from time series using single-cell datasets. It allows for simultaneous inference of sample-specific composition as well as overall time trajectory dynamics using highly scalable stochastic variational inference.

Install

Installation in an Anaconda environment is recommended.

Install pyro (https://pyro.ai/) and pytorch

Clone the repository

git clone git@github.com:broadinstitute/temporal-rna-seq-deconvolution
cd temporal-rna-seq-deconvolution

In ubuntu you might need to install the following for the packages to compile

sudo apt-get install libglib2.0-dev python3-dev

Install python package

pip install -e .

Documentation

Auto-generated code documentation for the package can be found on readthedocs.io.

Tutorials

For a full tutorial on the use of ternadecov please visit the GP deconvolution tutorial.

Additional tutorials provide information on:

• Simulating and saving datasets
• Simulating linear trajectories
• Simulating sigmoid trajectories
• Simulating periodic trajectories
• Evaluating the time required to run the deconvolution for different numbers of samples

CLI Use

ternadecov offers an cli for automated deconvolution and exporting of results from the command line. An example run call could be:

ternadecov deconvolve \
  --bulk-anndata bulk.h5ad \
  --sc-anndata singlecell.h5ad \
  --iterations 20000 \
  --sc-celltype-column Abbreviation \
  --bulk-time-column dpi_time \
  --feature-selection-method overdispersed_bulk_and_high_sc \
  --export-prefix results_ \
  --export-directory output/

For more information on CLI usage try:

ternadecov --help