Advanced Bayesian Modeling, MCMC Diagnostics, brms, Stan, and Piecewise SEM for Biological & Agricultural Sciences
Bayesian-Biostats is a comprehensive computational repository for Bayesian Data Analysis and Structural Equation Modeling in agronomy, plant biology, genomics, and ecology. It provides modular, publication-ready R and Stan scripts for fitting hierarchical models, evaluating MCMC convergence, selecting informative priors, and resolving direct/indirect causal pathways.
Datos Experimentales ──► Priors Informativos ──► MCMC (Stan/brms) ──► Diagnósticos (R-hat/ESS) ──► Inferencia & Piecewise SEM
(Agronomía/Bio) (Regularización) (Cadenas No-U-Turn) (bayesplot/LOO-CV) (Efectos Directos/Indirectos)
- Hierarchical split-plot, RCBD, and repeated measures experiments.
- Non-Gaussian likelihoods (Gamma, Beta, Negative Binomial, Zero-Inflated Poisson).
- Multi-level variance partitioning and shrinkage estimation.
- Direct and indirect effect decomposition for multi-trait physiological networks.
- Integration of mixed-effects sub-models within structural directed acyclic graphs (DAGs).
- Shipley's d-separation tests for directional independence and model fit.
- Calibrating weakly informative, regularizing, and domain-informed priors.
- Prior Predictive Checks (PPC) and sensitivity auditing.
- Trace plots, autocorrelation time, potential scale reduction factor (
$\hat{R} < 1.01$ ), and Effective Sample Size (ESS Bulk/Tail). - Out-of-sample predictive performance using Approximate Leave-One-Out Cross-Validation (LOO-CV) and Widely Applicable Information Criterion (WAIC).
- Core Engine:
R•brms•rstan•cmdstanr•Stan - Structural Models:
piecewiseSEM•lavaan•blavaan - Diagnostics & Posterior:
bayesplot•tidybayes•posterior•loo•DHARMa - Visualization:
ggplot2•ggdist•patchwork
Developed by Paúl Alexander López Peña • Pontificia Universidad Católica de Chile (PUC)