Optimal experiment design for cardinal values estimation: guidelines for data collection

Area of Science:

• Microbiology
• Statistics
• Experimental Design

Background:

• Current predictive microbiology models often use arbitrary or factorial experimental designs.
• These designs may not identify the most relevant experimental conditions for accurate parameter estimation.
• Optimal experiment design offers a statistically rigorous approach to improve model accuracy.

Purpose of the Study:

• To compute optimal experiment designs for parameter estimation (OED/PE) for the cardinal temperature model with inflection point (CTMI) and the cardinal pH model (CPM).
• To compare D-optimal and E-optimal designs for maximizing parameter estimation accuracy and confidence.
• To guide the selection of informative experimental conditions in predictive microbiology.

Main Methods:

• Utilized optimal experiment design principles, focusing on extremizing scalar functions of the Fisher information matrix.
• Performed model output sensitivity analysis to identify relevant temperature and pH ranges.
• Computed D-optimal designs (minimizing Fisher information matrix determinant) and E-optimal designs (maximizing smallest eigenvalue).

Main Results:

• Optimal designs typically select four informative temperature or pH levels, with replicate experiments at these points.
• Informative experiments are strategically placed at conditions with extreme model output sensitivity.
• Boundary constraints were necessary to exclude unworkable experiments and account for parameter uncertainties.

Conclusions:

• Optimal experiment design provides a superior alternative to arbitrary or factorial designs for secondary modeling in predictive microbiology.
• This approach significantly enhances parameter estimation accuracy and model reliability.
• The findings advocate for the use of OED/PE to optimize experimental efforts and improve microbiological models.