Modified frequency-based sensitivity analysis for oscillatory system models

Global sensitivity analysis (GSA) is a critical and widely used method for evaluating model performance, as it quantifies input variability and uncertainty in both linear and nonlinear systems, with or without parameter interactions. However, for quantitative systems pharmacology or systems biology models with complex dynamics and time-dependent output exhibiting a single dominant oscillatory frequency, the classic GSA computed on fixed-time outputs may overlook parameters controlling oscillatory behaviors. To address this, a Fourier-based approach was refined to extract a single dominant oscillatory frequency across model simulated temporal profiles and define them as the quantity of interest (QoI) to compute the Sobol indices. Surrogate modeling was used to improve computational efficiency, which reduced computation time by at least 10-fold while recovering the ranking of influential parameters. This methodology was applied to several cases, including one involving dexamethasone treatment, to showcase how drug-induced perturbations can alter parameter sensitivity. In conclusion, these cases demonstrate the utility and efficiency of a Fourier-based GSA for identifying parameters controlling a dominant frequency in oscillatory system models.