Structural identification of GMA models: algorithm and model comparison

TLDR: A local search algorithm for structural identification of Generalized Mass Action models from time course data is proposed and is able to find as good or better models than any of the other approaches.

Abstract: We propose a local search algorithm for structural identification of Generalized Mass Action (GMA) models from time course data. The algorithm has been implemented as part of our existing system for identification of dynamical systems.We compare this approach to existing alternatives in terms of analytical GMA models, analytical GMA models with parameter estimation from time course data, S-systems, and linear models. This is done on three new test problems designed to exhibit different characteristic properties of biochemical pathways, and which are defined with chemical rate reactions. By applying state-of-the-art algorithmic methods we are able to make a full investigation for the test problems also with noisy data.The results show that on the tested problems, our structural identification algorithm is able to find as good or better models than any of the other approaches. It can therefore be expected to be a useful tool for identifying models of unknown systems from time course data.All test problems are available on the web.