This Python package implements the Multi-Dendrix algorithm for analyzing driver pathways in cancer mutation data (PLoS Comp Bio, 2013). It was developed by the Raphael research group in the Center for Computational Molecular Biology at Brown University.
This package also includes a Python package that implements a Markov chain Monte Carlo (MCMC) version of Multi-Dendrix. Multi-Dendrix MCMC does not require IBM's CPLEX, and is thus completely open-source to use. We include a brief description of how to run Multi-Dendrix MCMC at the end of this README.
After pulling from Github, simply run:
python setup.py install
Multi-Dendrix was developed on 64-bit Debian Linux. It has not been tested on other systems.
The Multi-Dendrix package requires the following Python modules:
- [IBM's CPLEX] (http://goo.gl/dJV6f).
- [NetworkX] (http://networkx.github.com).
- Either SciPy >= version 0.11 or fisher0.1.4.
In addition, for web output with network figures, Multi-Dendrix requires the installation of GraphViz.
To run the full Multi-Dendrix pipeline on the provided data,
For other uses, please refer to our documentation (see below).
We offer extensive documentation for running Multi-Dendrix, as well as creating custom analysis pipelines, at http://raphael-group.github.com/multi-dendrix.
Please visit our Google Group to post questions and view discussions from other users.
The Multi-Dendrix MCMC algorithm samples collections of t gene sets of size k ( note that k is fixed, unlike in the Multi-Dendrix ILP) in proportion to their weight. The Multi-Dendrix release now includes a Python package where the MCMC algorithm is implemented as a Python C extension. To run Multi-Dendrix MCMC:
First, compile the C code to build the Multi-Dendrix MCMC Python module:
cd multi_dendrix_mcmc python compile.py build
Then, get a complete list of arguments to the Multi-Dendrix MCMC program by running:
cd ../ python run_multi_dendrix_mcmc.py -h
The output of Multi-Dendrix MCMC is a tab-separated file that lists the collections sampled by the MCMC algorithm. The first line of each collection lists the sampling frequency, weight W', permutation P-value (if computed), and the first gene set in the collection and its weight. The following t-1 lines list the remaining gene sets and their weights.
Multi-Dendrix MCMC computes the permutation P-values for the ith highest weight collection using the same procedure for permuting mutation data described in the Multi-Dendrix paper. The MCMC algorithm is run on each of these permuted datasets, and the weight of the ith highest weight collection from real data is compared to the weight of the ith highest weight collection across the permtued datasets. Multi-Dendrix MCMC only computes the permutation P-value for the top 25 highest weight collections from a given run.
We thank Troy D. Hanson for the uthash library used by the Multi-Dendrix MCMC C extension.