Convergence, sampling and total order estimator effects on parameter orthogonality in global sensitivity analysis.

Area of Science:

• Computational modeling
• Life sciences
• Sensitivity analysis

Background:

• Dynamical system models require sensitivity analysis to quantify input parameter effects and orthogonality.
• Prior research focused on estimator performance for parameter effects, but not parameter orthogonality for identifiability.

Purpose of the Study:

• Assess estimator and sampling methodology interactions for parameter orthogonality in cardiovascular models.
• Investigate convergence, dimensionality effects, and practical implications for life sciences modeling.

Main Methods:

• Utilized cardiovascular system numerical models.
• Evaluated various sampling methodologies and estimators, including Jansen, Janon, Sobol, and lattice rule.
• Performed resampling studies to assess sample convergence and impact of model dimensionality.

Main Results:

• Jansen or Janon estimators coupled with Sobol and lattice rule sampling demonstrated efficient convergence and minimal uncertainty.
• Global sensitivity analysis is convergence-dependent; unconverged indices lead to inaccurate parameter influence and orthogonality recovery.
• Interactions between estimators and sampling methods were clarified, reducing ambiguity.

Conclusions:

• The combination of Jansen/Janon estimators and Sobol/lattice rule sampling is recommended for robust parameter orthogonality and influence calculations.
• Ensuring convergence in global sensitivity analysis is critical for accurate parameter assessment.
• This study establishes new practices for dynamical system modeling in the life sciences by clarifying estimator-sampler interactions.