Model choice problems using approximate Bayesian computation with applications to pathogen transmission data sets.

Area of Science:

• Statistics
• Computational Biology
• Epidemiology

Background:

• Standard Bayesian inference struggles with intractable likelihoods in complex models.
• Approximate Bayesian Computation (ABC) methods use simulation but are less developed for model choice.
• Existing ABC methods face challenges in handling multiple model spaces.

Purpose of the Study:

• To develop an advanced ABC algorithm for statistical model choice.
• To extend existing ABC methods to effectively compare multiple competing models.
• To infer pathogen transmission models using a robust computational approach.

Main Methods:

• Developed an ABC algorithm incorporating stepwise multinomial logistic regression.
• Integrated reversible jump Markov chain Monte Carlo for enhanced model space exploration.
• Utilized posterior means of model parameters as summary statistics within a discrepancy measure.

Main Results:

• The algorithm demonstrated robustness across a range of true model probabilities in a validating example.
• Successfully applied to three pathogen transmission scenarios of varying complexity.
• Effectively inferred preferences for specific pathogen transmission models.

Conclusions:

• The proposed ABC algorithm provides a powerful tool for statistical model choice.
• The method is effective in selecting appropriate models for complex systems like pathogen transmission.
• Enhances the utility of ABC methods beyond parameter estimation to model selection.