mw201608 / BayesianNetwork

Bayesian network analysis pipeline

This repository provides a simple example pipeline to perform Bayesian probabilistic network inference using the RIMBANet software.

Usage:

1. Download folders bin, script, and Perls. Place them in a directory say $HOME/user/RIMBANet/. Add eXecute permission to files in the two folders. Lastly, either add $HOME/user/RIMBANet/bin to your system PATH variable or move bin/testBN to a place included in PATH.

2. Prepare discretized data matrix (discretized continuous values into ordinal 0, 1, and 2; see an example using kmeans clustering based discretization method) and save in a tab-delimited text file named data.discretized.txt in your working directory, say /path/to/working/directory. In this file, each row represents a variable, and each column represents a sample except the first columns which consists of variable names. However, there should be no header row.

3. Call RIMBANet to construct networks with different random starting structures. A wraper script caller.bsh is provided to create all required files from the data and then submit parallel jobs on a LSF environment. To change any of the default behabivors, please edit caller.bsh, BNJob.bsh, and BNJobbase.bsh. To add additional prior information, please refer to Perl scripts in folder Perls.

bash $HOME/user/RIMBANet/script/caller.bsh /path/to/working/directory

4. When LSF jobs (default 1000) are completed successfully, create a consensus network structure (default final output result.links3.links.txt) consists of de-looped links present in, by default, at least 30% of the 1000 individual structures.

bash $HOME/user/RIMBANet/script/createStructure.bsh bn.param.txt