Structured methods for parameter inference and uncertainty quantification for mechanistic models in the life...

Area of Science:

• Computational Biology
• Mathematical Modeling
• Scientific Computing

Background:

• Parameter inference and uncertainty quantification are crucial for linking mathematical models to real-world data.
• Current methods are often computationally intensive, especially with numerous model parameters.

Purpose of the Study:

• To develop and validate an efficient profile likelihood-based method for parameter inference.
• To leverage the inherent structure of mathematical models for computational efficiency.

Main Methods:

• Developed a profile likelihood method exploiting model structure, such as linear scaling parameters.
• Applied the method to diverse life science models: predator-prey, epidemic, and advection-diffusion reaction.
• Compared accuracy and computational cost against existing profile likelihood techniques.

Main Results:

• The new method achieves comparable accuracy to traditional profile likelihood approaches.
• Substantially fewer forward model evaluations are required, indicating significant computational savings.
• Demonstrated applicability across ecology, health, and environmental science models.

Conclusions:

• The proposed structured profile likelihood method offers a more efficient approach to parameter inference.
• This efficiency is particularly beneficial for models where structured parameter relationships can be identified.
• Publicly available code facilitates the application of this method to user-defined models and data.