Emulator-based Bayesian optimization for efficient multi-objective calibration of an individual-based model of...

Area of Science:

• Epidemiology
• Computational Biology
• Mathematical Modeling

Background:

• Individual-based models (IBMs) are crucial for infectious disease research but are difficult to calibrate.
• Model complexity hinders accurate fitting to real-world biological and epidemiological data.

Purpose of the Study:

• To present a novel Bayesian optimization framework for calibrating complex IBMs.
• To enhance the accuracy and interpretability of infectious disease transmission simulators.

Main Methods:

• Utilized a Bayesian optimization framework with Gaussian process or machine learning emulators.
• Optimized a complex malaria transmission simulator across a high-dimensional parameter space.
• Employed multiple fitting objectives derived from malaria natural history and disease progression data.

Main Results:

• The proposed approach rapidly outperformed previous calibration methods.
• Achieved a superior final goodness of fit for the malaria transmission model.
• Generated parameter importance and sensitivity diagnostics for enhanced interpretability.

Conclusions:

• Bayesian optimization offers an efficient and effective solution for calibrating complex IBMs.
• The method improves model predictive power and provides valuable epidemiological insights.
• This approach increases trust in model predictions through greater transparency and interpretability.