Enhancing Parameter Estimation of Biochemical Networks by Exponentially Scaled Search Steps

摘要

A fundamental problem of modelling in Systems Biology is to precisely characterise quantitative parameters, which are hard to measure experimentally. For this reason, it is common practise to estimate these parameter values, using evolutionary and other techniques, by fitting the model behaviour to given data. In this contribution, we extensively investigate the influence of exponentially scaled search steps on the performance of two evolutionary and one deterministic technique; namely CMA-Evolution Strategy, Differential Evolution, and the Hooke-Jeeves algorithm, respectively. We find that in most test cases, exponential scaling of search steps significantly improves the search performance for all three methods.