Pedigree-based mixed-effects models via DAGs

This workflow evaluates directed acyclic graph (DAG) formulations for the animal model using both JAGS and NIMBLE. It benchmarks Gibbs/Metropolis samplers, transformed additive-genetic parameterisations, and Cholesky-based approaches to understand how sampler choice affects effective sample size and CPU time.

Highlights

Demonstrates how transformed additive genetic values can improve mixing compared to classical Gibbs updates, while documenting the cost of dense Cholesky decompositions of the NRM.
Shows that NIMBLE consistently outperforms JAGS for complex animal models once custom samplers are defined.
Provides complete R scripts to install specific package versions (e.g., MasterBayes 2.58) and OS-specific build tools (Rtools, Xcode, GNU Fortran).

Required packages & environment

pacman::p_load(
  tidyverse, version = "1.3.2",
  knitr, version = "1.42",
  runjags, version = "2.2.1-7",
  MCMCvis, version = "0.15.5",
  MCMCglmm, version = "2.34",
  gdata, version = "2.18.0.1",
  MasterBayes, version = "2.58",
  coda, version = "0.19-4",
  ggrepel, version = "0.9.2",
  dplyr, version = "1.0.10",
  kableExtra, version = "1.3.4",
  MatrixModels, version = "0.5-1",
  pedigreemm, version = "0.3-3",
  prettycode, version = "1.1.0",
  formattable, version = "0.2.1",
  AlphaSimR, version = "1.3.4",
  patchwork, version = "1.1.2",
  animalModels
)

Additional dependencies: JAGS (https://sourceforge.net/projects/mcmc-jags/), platform-specific compilers (Rtools on Windows, Xcode/GNU Fortran on macOS, gfortran on Linux). The scripts show how to install archived versions when needed via remotes::install_version().

Usage notes

Clone the repo and follow the README instructions to install OS-specific toolchains and package versions.
Run the provided R Markdown files to compare samplers (MasterBayes, NIMBLE, JAGS) and inspect effective sample size, trace plots, and runtime summaries.
The outputs guide the choice of parameterisation/sampler for future animal-model analyses, especially when quantifying additive-genetic variance in breeding pipelines.