This repository contains the supplementary information accompanying the manuscript "Fides: Reliable Trust-Region Optimization for Parameter Estimation of Ordinary Differential Equation Models".

To set up the scripts, execute the script setup.sh, this will setup a virtual environment and download and install additional dependencies.

To ensure reproducibility, the scripts are organized using snakemake. To execute the python part of the benchmark, execute benchmarkLocal.sh . This will run the whole benchmark locally and likely need multiple days to finish. However, it should be easy to adapt this script to run on a cluster, which can finish in a couple of hours depending on available resources.

Running the python benchmark will generate results files in the ./results directory that include optimization results in hdf5 format as well as waterfall, parameter and convergence plots for every model and otimizer . Moreover additional evaluation figures will be generated in the ./evaluation directory.

The MATLAB part of this benchmark can be tun by executing the script ./Hass2019/run_Benchmark.m. The script will take a bit over a week to finish on modern hardware. The folder ./Hass2019 contains a modified version of the script arFit.m that ensures that all optimizer options are corrrectly applied to the fmincon optimizer. These changes will be automatically applied to the downloaded d2d version at ./Hass2019/d2d . This repository is already prepopulated with results from this optimization, which will be loaded instead of rerunning the benchmarks . To rerun a benchmark delete the respective .mat file in ./Hass2019

To compare results across optimizers and methods, the script comparison.py has to be executed after both MATLAB and python part of the benchmark have finished. This will generate additional figures and .csv files which serve as the basis for text and figure in the manuscript.