Fig. 1.

Tasks to be accomplished for fitting ODE models. The fitting of ODE models requires several generic tasks. The optimization problem has to be defined in terms of bounds of the search space and geometry (e.g., linear vs. log scale). Moreover, the selected generic optimization algorithm applied as the core of optimization-based fitting has to be initialized. There are many ways of combining global and local search strategies. A prominent global search strategy is random drawing of multiple initial guesses and performing local optimization for each starting point. In each optimization step of a local optimization run, the ODEs have to be solved for the evaluation of the objective function χ²(θ). Incremental improvement strategies are applied for suggesting a new parameter vector for the next iteration step in the core optimization routine which is usually performed based on derivatives or approximations thereof