GNW

C++ version of GeneNetWeaver

See also http://sourceforge.net/projects/gnw/

To simulate time series with constant input, add

GnwSettings::Instance()->generateTsConstantInput(true);

in main.cpp, otherwise keep this flag false.

Workflow

• Generate network topology by the generate_graph function in python/network.py, implemented with the NetworkX library. Within the script, one can specify the size of the network, the number of networks and the directory to save the edge list .tsv file. For example, start a python console by ipython within the python/ directory, then

  from network import generate_graph
  generate_graph('scalefree')
  

• Use GNW to generate the time series. To submit the job to the SGE queueing system, use the gnw.sh script, or for example

  for i in `seq 1 50`;do qsub -t 1-100 gnw.sh scalefree2 $i;done
  

  Here the -t option of qsub specifies the range of the array job, i.e. the same job is submitted in parallel with SGE_TASK_ID ranging from 1 to 100, which is the index of different network topologies. scalefree2 is the name of the network, and the second argument $i runs from 1 to 50, which is the index of different perturbations in this case.

  For the simulation of only one network, the equivalent is

  qsub -t 1 gnw.sh scalefree2 1
  

  Use watch qstat to monitor the progress of all jobs.

• Convert the GNW output to proper format. python/gnw_parser.py converts the time series (*multifactorial_timeseries.tsv) to format that is readable by the hitmds tool and save it in the mds/ subdirectory, to run in parallel

  qsub -t 1-100 python/fnet.sh scalefree2 50
  

  The array job range 1-100 defines the number of unique network topologies, and the last argument 50 is the number of different perturbations.

  In order to convert the sbml parameter file to a plain text file readable by R, use the python/param_loader.py script. One can open up the script in any text editor and go to the line with if __name__ == '__main__':, which is the main function. In general, the python script loops through all sbml files generated by different topologies and spits out 3 files (*_delta.csv, *_k.csv, *_n.csv) saving the decay rate (delta), dissociation concentration (k) and Hill coefficient (n) respectively.

• Run MDS. Code is located in the github/mds repository, see the documentation there or in a nutshell, one can submit the job to the queueing system by

  for i in `seq 1 100`;do qsub -t 1-50 mds.sh scalefree2 $i;done
  

  Notice that the syntax is similar to the gnw.sh script, but the network and perturbation index are interchanged, such that the array job index 1-50 denote different perturbations and the loop index 1-100 denote different topologies.

• Convert MDS coordinates to p-values with github/mds/R/response.R. Currently this R script cannot be submitted to the queueing system and therefore only runs on the head node. If one has multiple networks, one can use the bash script github/mds/R/response_loop.sh, which is basically a sequential loop.