No description, website, or topics provided.
C++ C Python Shell
Switch branches/tags
Nothing to show
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Parana 2 (Reviewer Pre-release)

Note: We're currently working migrating this project to Github and working on an easier-to-compile release which should be available shortly.

This software accompanies ISMB 2013 publication

"Predicting protein interactions via parsimonious network history inference"

and is for reviewer purposes only. It should not currently be redistributed. An initial release is forthcoming.


A tarball containing both the bZIP and Herpes virus data, already in the correct format and able to be processed by the program, can be obtained here.


The binary release should be ready to use (if you're on OSX).

The source version uses the CMake build system. It depends on a C++11 compatible (e.g. G++-4.7) compiler and the following external libraries:


From the top level directory, one should execute the following commands (> designates the prompt):

> mkdir build
> cd build
> cmake ..
> make
> mv bin ..

Any errors are likely the result of a missing library. The last line moves the binary directory (containing the parana2 executable) to the same relative path it's in in the binary release.

Running the program

The current pre-release version of the program exposes a number of options that are not fully implemented or which may be removed in the final version. However, all of the functionality necessary to reproduce the results of the paper are present and working. A typical invocation of the program looks something like this:

> ./parana2 pars -u -t [target network] -d [dup. hist] -o [output file]

The target network

The target network should be in the NetworkX adjacency list format. The comment character "#" is respected, and the file should list nodes with no incident edges on lines by themselves.

The duplication history

The duplication history should be in PhyloXML format (the NHX format will be supported shortly). Networks currently in the NHX format (specifically those that result from Notung) can be easily converted to NHX format using the phyloXML converter. For example, if one currently has a file "tree.ntg" (a NHX format tree that is the result of a gene-species tree reconciliation using Notung), she can obtain the appropriate phyloXML format file (assuming forester.jar is in the current directory) by running the command:

java -cp forester.jar org.forester.application.phyloxml_converter -f=nn tree.ntg tree.xml

This will output an appropriately formated phyloXML file, "tree.xml".

The output file

The output file is a list of posterior scores for all potential edges (for those with scores > 0). The format is very simple, each line is given as follows:

p1	p2	et	s

Where p1 and p2 are the names of two proteins, et is the edge-type (currently, this is just 'b'), and s is the score assigned to this edge.


Additional arguments are supported which affect the scores of different ancestral histories (e.g. the branch-length penalty and the ratio of cost of edge creation to deletion). These parameters are briefly documented with the program's help option (which can be viewd by running ./parana2 pars -h). The following parameters all currently work:

-r [--ratio]
-b [--beta] (this is called gamma in the paper --- will be fixed)
-k [--numOpt]
-p [--timePenalty]

Finally, to save the effort of rebuilding the hypergraph for every experiment when performing a cross-validation experiment (only the costs of the leaf hyperverties are affected in such experients), a cross validation flag (-c [--cross]) is available. This command modifies the meaning of the -o [--output] option, so that argument passed to this option will now become a directory, under which the results of all separate cross validation experiments will be written.


The binary and source release contain a number of scripts to assist in reproducing the results in the paper. First you should download the binary release, or download and compile the source release. Assume PDir is the top-level Parana2 directory (i.e. containing the data, scripts, etc. directories). Next, grab the data tarball above, or execute the following commands in PDir:

> cd data
> wget
> tar xzvf data.tgz

Now that the data is in the correct place, you can cd into the scripts directory.

> cd ../scripts

From here, there are shell scripts to run the ancestral network reconstruction at different noise levels:

> sh -d ../data -o ../output/noisetest

the bZIP cross-validation experiments,

> sh -d ../data -o ../output/bzipcross

and the Herpes cross-validation experiments,

> sh -d ../data -o ../output/herpescross

In addition to numerous other scripts, the scripts directory also includes the scripts that can analyze these results determine the algorithm's performance. In particular, there are 3 scripts that are relevant:

  • --- Analyzes the result of an network history inference.
  • --- Evaluates cross-validation experiments on the bZIP network
  • --- Evaluates cross-validation experiments on the Herpes networks

Each of these scripts is a python script and can be run with the -h option to document it's input.

The Python scripts, themselves, rely on some libraries. In particular, to run any of the scripts you'll need docopt. Any of the scripts that analyze the phylogeny data will need NetworkX to deal with the graphs and ete2 to deal with the phyloXML phylogenies. Additionally, for computation and plotting, they use NumPy, SciPy and Matplotlib. Finally, to compute the BEDROC scores (in the AnalyzePredictions script), we use the Croc libraries. Despite the long list, most of these libraries are fairly standard, and you should be able to install them all via easy_install or pip.