Compound optimal design criteria for nonlinear models.

Area of Science:

• Computational Statistics
• Mathematical Modeling
• Chemical Engineering (Pharmacokinetics)

Background:

• Parameter estimation in nonlinear models often conflicts with design criteria.
• Existing methods for combining these objectives have limitations.
• Optimizing models requires balancing predictive accuracy with specific design goals.

Purpose of the Study:

• To propose and evaluate novel approaches for integrating parameter estimation with opposing design criteria in nonlinear models.
• To compare these new methods against a reference technique from the literature.
• To demonstrate the applicability of the methods using pharmacokinetic (PK) and generalized linear models.

Main Methods:

• A reference method maximizing a weighted product of efficiencies.
• A second method maximizing an opposing criterion while minimizing a loss function.
• A third method using a multi-objective simulated annealing algorithm to simultaneously optimize parameter estimation efficiencies and an opposing criterion.

Main Results:

• The study presents three distinct methodologies for addressing the challenge of combined parameter estimation and opposing design criteria.
• The simulated annealing approach offers a simultaneous optimization strategy.
• The methods are validated using established PK and generalized linear models from scientific literature.

Conclusions:

• The developed approaches provide effective strategies for optimizing nonlinear models with competing objectives.
• The multi-objective simulated annealing algorithm presents a powerful tool for simultaneous optimization.
• These methods enhance the design and analysis of nonlinear models in various scientific fields.