Animal models and integrated nested Laplace approximations.

Area of Science:

• Evolutionary Biology
• Animal Breeding
• Quantitative Genetics
• Statistical Genetics

Background:

• Animal models are crucial for dissecting the genetic basis of traits in evolutionary biology and animal breeding.
• Bayesian inference for these complex hierarchical models traditionally relies on computationally intensive Markov Chain Monte Carlo (MCMC) methods.
• Integrated Nested Laplace Approximation (INLA) provides a faster, non-sampling alternative for Bayesian inference in Gaussian Markov random field models.

Purpose of the Study:

• To demonstrate the applicability and efficiency of the INLA methodology for a wide range of Bayesian animal models.
• To compare the performance of INLA against MCMC methods for inference in animal models.
• To introduce the "AnimalINLA" R package for facilitating rapid Bayesian animal model inference.

Main Methods:

• Application of INLA to synthetic and real-world datasets (house sparrow populations).
• Analysis of animal models with Gaussian, binomial, and Poisson likelihoods.
• Comparison of INLA results with MCMC inference and model selection using the Deviance Information Criterion (DIC).

Main Results:

• INLA successfully performs Bayesian inference for diverse animal models across different likelihoods.
• Inference results from INLA are comparable to those obtained using MCMC methods.
• The study provides a framework for evaluating differences in DIC, aiding model selection.

Conclusions:

• INLA is a viable and efficient alternative to MCMC for Bayesian animal models, accelerating genetic analyses.
• The "AnimalINLA" package simplifies the implementation of INLA for researchers in evolutionary biology and animal breeding.
• This approach enhances the speed and accessibility of genetic trait analysis in animal populations.