Approximate Bayesian computation for spatial SEIR(S) epidemic models.

Area of Science:

• Computational Statistics
• Epidemiology
• Computational Biology

Background:

• Bayesian inference is crucial for complex models but often computationally intractable.
• Traditional Markov chain Monte Carlo (MCMC) methods become infeasible for large-scale or spatially explicit models.
• Approximate Bayesian Computation (ABC) offers a parallelizable alternative for such problems.

Purpose of the Study:

• To demonstrate the application of Approximate Bayesian Computation (ABC) for estimating parameters in spatially heterogeneous Susceptible-Exposed-Infectious-Removed (SEIR) epidemic models.
• To introduce the open-source ABSEIR package in R for practical implementation of ABC in epidemic modeling.
• To compare the performance of ABC against MCMC methods in both simulated and real-world epidemic scenarios.

Main Methods:

• Application of Approximate Bayesian Computation (ABC) algorithms.
• Development and utilization of the open-source ABSEIR package for R.
• Simulation studies comparing ABC and MCMC performance.
• Analysis of the 2014 Chikungunya epidemic in the Americas using spatially heterogeneous SEIR models.

Main Results:

• ABC successfully estimates parameters in complex, spatially heterogeneous SEIR models where MCMC is computationally infeasible.
• The ABSEIR package provides a practical tool for implementing these ABC techniques.
• Performance comparisons show ABC is a viable and often more efficient alternative to MCMC for these models.

Conclusions:

• Approximate Bayesian Computation (ABC) is a powerful and practical approach for Bayesian inference in complex epidemic models.
• The ABSEIR R package facilitates the application of ABC to real-world epidemiological challenges.
• ABC methods are essential for advancing our ability to model and understand disease spread in heterogeneous populations.