Evolving Improved Sampling Protocols for Dose-Response Modelling Using Genetic Algorithms with a Profile-Likelihood...

Area of Science:

• Pharmacokinetics and Pharmacodynamics (PK-PD) modeling
• Computational biology
• Experimental design optimization

Background:

• Mathematical models are crucial for understanding biological systems, but data limitations can hinder precise parameter identification.
• Existing model-based experimental design methods often rely on local approximations, potentially underestimating parameter uncertainty.
• Profile-likelihood methods offer a more robust way to quantify parameter uncertainty beyond linear assumptions.

Purpose of the Study:

• To develop and evaluate a genetic algorithm (GA) approach for optimizing sampling times in PK-PD models.
• To minimize parameter uncertainty using a profile-likelihood-based metric.
• To address simultaneous optimization for multiple parameter scenarios in experimental design.

Main Methods:

• A genetic algorithm was employed to optimize sampling schedules for a PK-PD model.
• The optimization objective was based on a profile-likelihood derived metric for parameter uncertainty.
• The GA approach was tested across various sample numbers (n=3-20) and parameterizations.

Main Results:

• The GA successfully identified near-optimal sampling protocols, reducing parameter variance by 33-37% on average.
• The profile-likelihood metric showed strong correlation (r > 0.89) with a Monte Carlo-based metric.
• Computational cost was reduced by an order of magnitude compared to existing methods.

Conclusions:

• The combination of a GA and profile-likelihood metric enables consideration of model nonlinearity in experimental design.
• This approach offers a feasible and computationally efficient method to improve parameter certainty or reduce sample size.
• Optimized experimental design can lead to more precise parameter identification and efficient resource allocation in biological research.