Sloppy models can be identifiable

Abstract: Dynamic models of biochemical networks typically consist of sets of non-linear ordinary differential equations involving states (concentrations or amounts of the components of the network) and parameters describing the reaction kinetics. Unfortunately, in most cases the parameters are completely unknown or only rough estimates of their values are available. Therefore, their values must be estimated from time-series experimental data. In recent years, it has been suggested that dynamic systems biology models are universally sloppy so their parameters cannot be uniquely estimated. In this work, we re-examine this concept, establishing links with the notions of identifiability and experimental design. Further, considering a set of examples, we address the following fundamental questions: i) is sloppiness inherent to model structure?; ii) is sloppiness influenced by experimental data or noise?; iii) does sloppiness mean that parameters cannot be identified?, and iv) can sloppiness be modified by experimental design? Our results indicate that sloppiness is not equivalent to lack of structural or practical identifiability (although they can be related), so sloppy models can be identifiable. Therefore, drawing conclusions about the possibility of estimating unique parameter values by sloppiness analysis can be misleading. Checking structural and practical identifiability analyses is a better approach to asses the uniqueness and confidence in parameter estimation.