View source for Hybrid optimization method with general switching strategy for parameter estimation

__NUMBEREDHEADINGS__
=== Citation ===
Eva Balsa-Canto, Martin Peifer, Julio R Banga, Jens Timmer and Christian Fleck 
[https://doi.org/10.1186/1752-0509-2-26 Hybrid optimization method with general switching strategy for parameter estimation], 2008,
BMC Systems Biology, 2:26.

=== Summary ===
Parameter estimation in systems biology model requires finding the global optimum of the objective function. This is usually done in terms of deterministic local optimizers which run the risk of precisely finding only a local optimum or by stochastic global optimizers which find the global optimum, but do not fully converge. Hence, the authors proposed a hybrid optimization which switches from stochastic global optimization to a local ''multiple shooting'' approach with a switching point which is determined dynamically. The hybriod method was demonstrated to vastly outperform stochastic global optimization and single as well as multiple shooting local methods in terms of computation time.

=== Study outcomes ===

==== Multiple vs Single Shooting ====
As a deterministic optimizer, the multiple shooting method is better at finding the global optimum than a single shooting approach, as can be seen from Figure 1 and 2 in the original publication. However, this comes at the cost of increased computational cost.

==== Comparison of Global Optimizers ====
A comparison of single shooting, multiple shooting, a stochastic global optimizer and the hybrid method showed that the hybrid method decreased the computation time significantly compared to the alternatives in the two tested models, which is listed in Table 1 and Table 2 and demonstrated in Figure 3 in the publication.

=== Study design and evidence level ===
*The 2 tested models although realistic, have a rather small parameter space (less than 10 parameters).
*The used data was simulated. Furthermore, noise levels are rather mild (0% and 10% noise to signal ratio).
*Parameter bounds from which initial guesses for local fitting were drawn are rather narrow. Nevertheless, multiple shooting fails a considerable amount of times for the widest choice of initial conditions.

=== Further comments and aspects ===
*The authors claim that multistart methods are outdated, yet such a method was reported to be superior to competing algorithms in the benchmark study [[Lessons Learned from Quantitative Dynamical Modeling in Systems Biology]].