Better than Random analysis

Authors: Petar Todorov¹, Artem Sokolov^1
1Laboratory of Systems Pharmacology, Harvard Medical School

What is this?

This repo aims to offer an easy way to test gene set hypotheses against backgrounds of randomly chosen gene sets predictions. Gene sets which are more predictive than the distribution of randomly chosen gene sets may indicate a link to the importance of their constituent genes.

Prerequisites

If you are running on a Linux system which has multiple versions of python, gcc, and git you should go ahead and make sure you have the correct ones by running which <module name>. To discover what version of these modules are available use module avail <module name> and then module load <module name> to load the correct one before proceeding. The btr package has been tested with with python/3.6.0, gcc/6.2.0, and git/2.14.2 or higher.

We get started by creating a Python 3 virtual environment.

virtualenv nameyourenvhere

In case you are setting this up on a cluster, you may want to use the packages compiled by the cluster. To do that:

virtualenv nameyourenvhere --system-site-packages

To activate your environment

source nameyourenvhere/bin/activate

Then clone this repo, and install it as editable using pip

git clone https://github.com/pvtodorov/btr.git
cd btr
pip install -e .

You're ready to go!

Using the software

Installing the repo will also bind some commands that can be used in the terminal. In order to specify how to run the software, a settings file is needed. An example can be see in this repo's example_settings.json. If a background is being generated, a file such as example_background_params.json must be provided. If a hypothesis is being used as the feature set, a .gmt file must be used such as this this from Pathway Commons.

Making predictions

To generate background predictions:

btr-predict <path to settings file> -b <path to background parmeters>

To generate geneset predictions:

btr-predict <path to settings file> -g <path to GMT file or folder with txt gene lists>

Evaluating predictions

To evaluate predictions, first score the background runs

btr-score <path to settings file>

To evaluate gene files, score them

btr-score <path to settings file> -g <path to GMT file or folder with txt gene lists>

Testing for statistical significance

btr-stats <path to settings file> -g <path to GMT file or folder with txt gene lists>