Mitra2019
This folder contains 31 example fitting problems configured for PyBioNetFit.
- problems.xls - Summary of all included fitting problems
Each folder contains the following files associated with a single fitting problem:
- README.rtf containing a detailed description of the fitting problem. This file also contains information about how the fitting job was run (under the heading “Fitting Run”) and performace of the fitting run (under the heading “Run times”)
- Model file(s) (extension .bngl or .xml) used as input for PyBioNetFit. These files have names specific to the problem.
- Data file(s) (extension .exp or .prop) used as input for PyBioNetFit. These files also have names specific to the problem
- Configuration files for PyBioNetFit. One file is provided for each algorithm tested. These files have names ending in “-de.conf”, “-ade.conf”, “-pso.conf”, and “-ss.conf”, indicating which algorithm is used by each file.
- For models fit to synthetic data, the ground truth model, with name ending in “_ground”
- Folders fit_ade, fit_de, fit_pso, and fit_ss containing example output from PyBioNetFit running each of the four algorithms. Each of these folders contains a list of the best parameters found (sorted_params_final.txt) and model file(s) generated by PyBioNetFit containing the best-fit parameters. If fitting was not attempted with a particular algorithm, the corresponding output folder is not included. If fitting was attempted but failed to reach the target objective function value, the output folder contains the best parameters at the time we terminated the algorithm.
Additional analysis was done on some of the examples. Such analysis is noted in problems.xls. For these problems, additional output files are provided describing the results of these analysis. Such files are described in the individual problem README files.
This folder will be included as Supplementary Information in Mitra, Suderman, et al. "PyBioNetFit and the Biological Property Specification Language" (in review for iScience).
PyBioNetFit is available at http://github.com/lanl/pybnf.
In most cases, fitting was run using PyBioNetFit v0.2.2 with four algorithms on default settings: differential evolution, asynchronous differential evolution, particle swarm, and scatter search. Fitting was run until a target objective was reached, with the target determined from a known ground truth (if data was synthetic), or from a previously reported fit for the problem. These are not the only possible choices for a target objective function value, but were made intending to achieve a good trade-off between performance and CPU time. For some problems, our chosen target is 5% above a previously reported best fit; this threshold of 5% is arbitrary. If a lower objective value is desired for a particular example, the fit can be rerun after modifying the min_objective configuration key of that example.