BLUPF90

BLUPF90 performs mixed-model computations to estimate variance components via residual maximum likelihood (REML) and Bayesian Gibbs sampling and to predict genetic values using best linear unbiased prediction (BLUP) for animal breeding applications.

Key Features:

• Mixed Model Computations: Handles mixed models with fixed and random effects to analyze phenotypic data for genetic merit prediction.
• Variance Components Estimation: Implements residual maximum likelihood (REML) methods and Bayesian approaches using Gibbs sampling for estimating variance components.
• Algorithm Selection Guidelines: Provides guidance on choosing between REML and Bayesian methods based on dataset characteristics and model complexity.
• Stability and Convergence: Addresses convergence issues, discusses Expectation-Maximization (EM) REML stability for singular covariance matrices, and highlights challenges with random regression models.
• Computational Efficiency: Incorporates computational optimizations to improve processing speed and handle large datasets and complex models.
• Flexibility in Model Complexity: Supports a range of model complexities and provides solutions for general models where REML may become unstable as trait number increases.

Scientific Applications:

• Animal Breeding Genetic Evaluation: Predicts genetic values and supports selection and improvement programs through mixed-model analyses.
• Quantitative Genetics Research: Enables estimation of variance components and investigation of genetic architecture using REML and Bayesian methods.

Methodology:

Implemented in Fortran 90/95 and employing mixed models with BLUP, residual maximum likelihood (REML) including EM-REML, Bayesian inference via Gibbs sampling, random regression models, handling of singular covariance matrices, and computational optimizations.