Graphical models and Bayesian domains in risk modelling: application in microbiological risk assessment

Area of Science:

• Food Safety and Microbiology
• Computational Statistics
• Risk Assessment Modeling

Background:

• Quantitative microbiological risk assessment (QMRA) models analyze hazard propagation in food chains.
• Monte Carlo (MC) simulation is standard for probabilistic risk assessment, aiming to understand parameter interdependencies and risk mitigation effects.
• Assessing the reliability of QMRA conclusions under uncertainty is crucial.

Purpose of the Study:

• To explore the application of Bayesian computational statistics within QMRA.
• To compare MC modeling with Bayesian inference (BI) and highlight limitations of MC.
• To demonstrate how BI can enhance MC models and formulate QMRA as Bayesian graphical models (BGMs).

Main Methods:

• Elaboration on the application of Bayesian computational statistics in QMRA.
• Exploration of the analogy between MC modeling and BI, particularly for prior distributions.
• Integration of BI submodels ('Bayes domains') into MC models and formulation of QMRA as BGMs.

Main Results:

• Identified limitations of MC modeling in handling feedback among model parameters.
• Demonstrated that integrating BI models into MC simulations overcomes these limitations.
• Showcased the formulation of entire QMRA models as BGMs, highlighting similarities with MC and BI.

Conclusions:

• Bayesian inference offers a powerful approach to enhance QMRA by addressing parameter feedback.
• Integrating BI into MC models ('Bayes domains') improves the reliability of risk assessment conclusions.
• Formulating QMRA as Bayesian graphical models provides advantages for understanding complex interdependencies.