ReactionNetworkEvolution.jl

This module allows the user to easily run an evolutionary algorithm to create mass-action chemical reaction networks with specific behaviors or to match time series data. It is intended as a research tool for computational systems biology.

There are three primary use cases for this packge:

1. Evolving reaction networks with specific behaviors. The default is oscillation, but other behaviors may be added in the future. Additionally, users can input time series that reflect those desired behaviors.
2. Evolving reaction networks (or ensemble models) to fit time series data. Often systems biologist will have timeseries data from experiments and will want to search for a reaction network (or group of them) that can recapitulate the time series data. This is useful for building models, informing future experiments or looking at output models for common reaction themes which may indicate the presence of those reactions in the ground truth.
3. Parameter (rate constants) fitting. In the case where model topology is known (for example, the biologist has already built the model from first principles) and wants to generate parameter values for those reactions that will recapitulate experimental time series data.

This module uses the julia programming language. Installation instructions can be found here.

1. Download the package
   From a julia console:

   using Pkg
   Pkg.add("ReactionNetworkEvolution")
   

2. Run evolution with default settings Remaining in the julia console:

   import ReactionNetworkEvolution as rne
   rne.run_evolution()
   

   NOTE: If an output directory is not supplied, then one will be created in the current working directory. Make sure that the working directory allows new directories to be made.

3. Customize Evolution (optional) The run_evolution() function takes several optional keyword arguments:

• ntrials: the number of trials to run (100 by default)
• ngenerations: the number of generations per batch (800 by default)
• population_size: the number of networks in a population
• pathtosettings: path to a json file storing additional custom settings
• outputpath: path to a directory where evolution output will be written
• seed: specify a seed for the random number generator

For example, to change the population_size to 200 via the command line:
rne.run_evolution(population_size=200)

Additional settings can be specified in a JSON file and supplied to the run_evolution() function as a keyword argument.
run_evolution(pathtosettings="/home/name/path/to/your/settings.json")

An example of a JSON file specifying custom settings:

{"chemical_species_names": ["A", "B", "C"],
"initial_concentrations": [5.0, 3.0, 6.0],
"reaction_probabilities": [0.2, 0.3, 0.3, 0.2],
"ngenerations": 500,
"population_size": 200,
}

Any settings that are not specified in the JSON file will be set to the default value.

This package was primarily designed to generate mass-action chemical reaction networks with oscillatory behavior. With the defualt settings, it will generate oscillators with 3 chemical species.

1. Download the package
   From a julia console:

   using Pkg
   Pkg.add("ReactionNetworkEvolution")
   

2. Run evolution with default settings Remaining in the julia console:

   import ReactionNetworkEvolution as rne
   rne.run_evolution()
   

1. Supply the time series data as a CSV file.
2. Create a JSON file with at least the following setting to direct the package to the time series CSV and disable sorting oscillators

{"objective_data_path": "/path/to/timeseries_data.csv",
"process_output_oscillators": false}

3. Run evolution supplying the JSON file:

  import ReactionNetworkEvolution as rne
  rne.run_evolution(pathtosettings="/path/to/settingsfile.json")

1. Supply the time series data as a CSV file
2. Supply the network topology as an antimony file.
3. Create a JSON file with at least the following settings:

{"objective_data_path": "/path/to/timeseries_data.csv",
"process_output_oscillators": false,
"seed_network_path": "/path/to/antimonyfile.txt",
"rateconstant_distance_weight": 1.0,
"p_rateconstant_mutation" 1.0}

The rateconstant_distance_weight setting enables the package to consider differences in reaction weight constants when comparing twp networks. This is necessary if all networks have the same topology, although some adjustments to the value may be necessary.

Setting p_rateconstant_mutation to 1.0 prevents the algorithm from adding or deleting reactions during evolution.

4. Run evolution supplying the JSON file:

  import ReactionNetworkEvolution as rne
  rne.run_evolution(pathtosettings="/path/to/settingsfile.json")