BranchedGP

BranchedGP is a package for building Branching Gaussian process models in python, using TensorFlow and GPFlow. The model is described in the paper

"BGP: Branched Gaussian processes for identifying gene-specific branching dynamics in single cell data", Alexis Boukouvalas, James Hensman, Magnus Rattray, bioRxiv, 2017..

This is now published in Genome Biology.

Example

An example of what the model can provide is shown below.

1. The posterior cell assignment is shown in top subpanel: each cell is assigned a probability of belonging to a branch.
2. In the bottom subpanel the posterior branching time is shown: the probability of branching at a particular pseudotime.

Install

If you have any problems with installation see the script at the bottom of the page for a detailed setup guide from a new python environment.

• Install tensorflow

pip install tensorflow

• Install GPflow

git clone https://github.com/GPflow/GPflow.git
cd GPflow    
pip install .
cd

See GPFlow page for more detailed instructions.

• Install Branched GP package

git clone https://github.com/ManchesterBioinference/BranchedGP
cd BranchedGP
python setup.py install
cd

Quick start

For a quick introduction see the notebooks/Hematopoiesis.ipynb notebook. Therein we demonstrate how to fit the model and compute the log Bayes factor for two genes.

The Bayes factor in particular is calculated by calling CalculateBranchingEvidence after fitting the model using FitModel.

This notebook should take a total of 6 minutes to run.

Tests

To run the tests should takes < 3min.

pip install nose
pip install nose-timer
cd BranchedGP/testing
nosetests --pdb-failures --pdb --with-timer

List of notebooks

To run the notebooks

cd BranchedGP/notebooks
jupyter notebook

Comparison to monocle-BEAM

In the paper we compare the BGP model to the BEAM method proposed in monocle 2. In monocle/runMonocle.R the R script for performing Monocle and BEAM on the hematopoiesis data is included.

List of python library files

Running in a cluster

When running BranchingGP in a cluster it may be useful to constrain the number of cores used. To do this insert this code at the beginning of your script.

from gpflow import settings
settings.session.intra_op_parallelism_threads = NUMCORES
settings.session.inter_op_parallelism_threads = NUMCORES