Benchmarks for identification of ordinary differential equations from time series data
We here present a collection of benchmark problems for ordinary differential equation (ODE) system identification from time series data. We consider problems where both structure (the form of the ODEs) and parameters are unknown. The problems are all related to biological applications and consist of a combination of problems that have been collected and translated from the biological literature, as well as a number of newly designed problems. However, the way problems are defined is not application specific and can be used in many areas.
The identification problems are defined mathematically as optimization problems, and are represented in a file format designed for this purpose. The format makes it easy to define and exchange identification problems, and a problem is fully defined in a single file. Please note that since these problems are optimization problems and not models, it is not possible to represent such problems in SBML. See the Introduction for more information.
The purpose of the collection is to facilitate evaluation of identification algorithms during development, and also and enable comparisons between different algorithms. We provide not just the problems, but also the best known solution of every problem, as well as any known source model. A NEW FEATURE is the possibility to verify identification problem files online, and to solve identification problems online by running on our server.
Detailed documentation and supplementary software is available in the Documentation. The official publication presenting the collection is Gennemark and Wedelin, Benchmarks for identification of ordinary differential equations from time series data, Bioinformatics 25(6):780-6.
In a new paper we consider parameter estimation only (Gennemark and Wedelin, Improved Parameter Estimation for Completely Observed Ordinary Differential Equations with Application to Biological Systems. P. Degano and R. Gorrieri (Eds.): CMSB 2009, LNBI 5688, pp. 205-217, 2009. Springer-Verlag Berlin Heidelberg). The considered problems are available here as well.
Please contact us if you wish to submit a problem to the collection, or if you have found a better solution to an existing problem! We are also interested in feedback on how to best define and represent identification problems.
| Peter Gennemark |  |
| Dag Wedelin |  |