Skip to content
Generation of flow networks robust against damages.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
classes
headers
CMakeLists.txt
LICENSE.rst
README.rst
rfn_config.hpp.in
simulation.cpp

README.rst

Flow Network Generation

Flow networks robust against damages are simple model networks described in a series of publications by Kaluza et al.[1, 2, 3]. The C++ code presented in this repository allows for the generation of such networks via a simulated evolution.

Although the code was programmed with a single core in mind, the compiled binary can easily be (and has been) run multiple times in parallel.

Installation

In the main directory, you need to edit the CMakeLists.txt file. If your libraries are installed in non-standard locations, please adapt lines 8 to 11. Then run the following commands:

cmake init .
make

If everything went smoothly there should now be a bin and lib subdirectory. Add the path to your environment variable LD_LIBRARY_PATH if you want to run things from here. In bash this can be done by:

export LD_LIBRARY_PATH="$HOME/path/to/rfn-generation/lib:$LD_LIBRARY_PATH"

You can now run the simulation binary in the bin subdirectory which will print some info to stdout.

If you want to install the bin and lib subdirectories in a different location, either edit line 17 of the CMakeLists.txt file or invoke:

make install -DDESTDIR=/your/favourite/path

When you are ready to move from testing to large-scale computation you should rebuild the project without debugging and text output, follow these commands:

cmake -DDEBUG=OFF .
make

Enjoy!

Note

The output files are binary and their exact structure depends on your system's architecture (32 or 64 bit).

Requirements

C++:

NB: If you install these libraries from system packages, please make sure to also install the dev packages as the headers are needed for compilation.

Others:

References

[1]Kaluza, P., Ipsen, M., Vingron, M. & Mikhailov, A. S. Design and statistical properties of robust functional networks: A model study of biological signal transduction. Physical Review E 75, 15101 (2007).
[2]Kaluza, P. & Mikhailov, A. S. Evolutionary design of functional networks robust against noise. Europhysics Letters 79, 48001 (2007).
[3]Kaluza, P., Vingron, M. & Mikhailov, A. S. Self-correcting networks: function, robustness, and motif distributions in biological signal processing. Chaos 18, 026113 (2008).
You can’t perform that action at this time.