The estimation of model parameters is the key function of the compartmental analysis task. In PKanalix it is performed in the framework of generalized least squares estimation. The optimization algorithm consists in minimizing a cost function with an improved version of the Nelder-Mead simplex algorithm, starting from multiple guesses in addition to the initial values provided. To run the algorithm click on the button COMPARTMENTAL ANALYSIS in the RUN tab.
The goal of the optimization is to find model parameters that give the best model fit to the observed data – minimize the objective (cost) function.
- The optimization algorithm performs well in general. However, even if it tries to explore the parameter space, the core algorithm (Nelder-Mead) is local. Therefore proper initialization of parameters is crucial.
- The cost function is described in this documentation page.
- In case of censored data, you can choose a method to treat BLQ data during the optimization: missing, equal to zero, equal to LOQ or equal to LOQ/2.
- By default, the optimization is performed for each individual independently. Switching on the toggle “Pooled fit” in the calculation settings will fit the same parameters to all individuals (data from all individuals is analysed together).
- If occasions are present and pooled fit is not selected, parameters are all considered occasion-dependent so the optimization for every individual includes \( N_{param} \times N_{occ}\) parameters (with \( N_{param} \) the number of parameters in the structural model and \(N_{occ}\) the number of occasions for this individual).
