Andrés Aravena, Dec 17, 2014
Combines a putative transcriptional regulatory network (such as those predicted using MEME) and co-expression data (such as microarray experiments results) to produce a realistic transcriptional regulatory network.
Lombarde models transcriptional regulatory networks in a scheme that integrates putative transcriptional regulatory networks with co-expression data to determine the simplest and most confident sub-network that explains the observed co-expressions.
The output will be a subgraph of the putative transcriptional regulatory network that satisfies the Lombarde criteria: each pair of co-expressed vertices should share a common regulator (either direct or via a regulation cascade), among all the common regulators select the most confident ones.
The Lombarde model is directly implemented on the lombarde.R script, whose inputs are graphs and whose outputs are a new graph and optionally a log on how this new graph is produced. This is the fists tool provided.
These inputs graphs are derived from experimental data that is preprocessed by standard tools like MEME/FIMO, BLAST and MRNET, and by ad-hoc scripts such as build_fimo_blast_net.py, discretize-weight.R and contract.R.
For a first approach to this suite of tools we also provide a “all-in-one” tool, named lombarde-full.sh. This is essentially a wrapper to all the ad-hoc scripts so all preprocessing is carried on automatically.
Currently most of the files represent graphs in NCOL format as parsed by igraph library on R (http://igraph.org/r/doc/read.graph.html).
The basic lombarde.R tool is then invoked as:
lombarde.R -o output.ncol -a output.log coexp.ncol putativeTRN.ncol
where coexp.ncol represent the set of co-expressed elements, one pair per line, and putativeTRN.ncol represents the putative transcriptional network built based on the output of BLAST and MEME/FIMO.
This last input file can be built doing:
build_fimo_blast_net.py fimo.txt blastp.txt coupling.txt > putativeTRN.txt