MultiEnvGS-Demo

This repository contains practical R scripts to demonstrate genomic selection models across multiple environments, with a focus on applications in plant breeding education and training.

• Hands-On:

What's Included (English)

1. Cross-validation functions: Scripts implementing different cross-validation schemes (CV0, CV00, CV1, and CV2) commonly used in genomic selection.
2. Bilingual instructions: Scripts with usage instructions available in both Portuguese (GS_Demo_ptbr.R) and English (GS_Demo_eng.R), including modeling code that uses the sommer package to perform multi-environment genomic selection.
3. Example datasets: Phenotypic (pheno_data.csv), genomic (Markers.Rda), and environmental data (W_matrix_USP_248) provided via https://github.com/gcostaneto/KernelMethods.

Model specification: The models used follow the structure:
y = u + e + g + g×e

Where:

• y: phenotype
• u: intercept
• e: environmental effect or environmental covariates effects
• g: genotypic or additive genomic effect
• g×e: genotype-by-environment interaction

O que está incluído (Português)

1. Funções de validação cruzada: Scripts que implementam diferentes esquemas de validação cruzada (CV0, CV00, CV1 e CV2), comumente usados na seleção genômica.
2. Instruções bilíngues: Scripts com instruções de uso disponíveis em português (GS_Demo_ptbr.R) e inglês (GS_Demo_eng.R), incluindo código de modelagem que utiliza o pacote sommer para realizar seleção genômica multiambiente.
3. Conjuntos de dados de exemplo: Dados fenotípicos (pheno_data.csv), genômicos (Markers.Rda) e ambientais (W_matrix_USP_248) disponíveis em https://github.com/gcostaneto/KernelMethods

Especificação do modelo: Os modelos seguem a estrutura:
y = u + e + g + g×e

Onde:

• y: fenótipo
• u: intercepto
• e: efeito do ambiente ou efeitos de covariáveis ambientais
• g: efeito genotípico ou efeito genômico aditivo
• g×e: interação genótipo × ambiente

Pipeline

Functions

fold

fold(fold.n, n)

• fold.n: fold.n: number of desired partitions (folds) for cross-validation. Should be ≥ 2.
• n: total number of observations to be split into folds.

Helper function that returns a list of indices for cross-validation splits. Each list element contains the indices of observations assigned to one fold. Used to divide data into roughly equal-sized subsets.

mmes.fold

mmes.fold(dados, G, fold)

• dados: data frame with phenotypic data containing 'gid' (genotype) and 'value' (phenotype) columns.
• G: genomic additive relationship matrix (e.g., GBLUP matrix), with row and column names matching 'gid' levels.
• fold: list of partitions generated by the fold() function.

Performs cross-validation on single-environment data. For each fold, the function fits a mixed model with mmes(), masking phenotypes in the test fold. Returns predictive accuracy as the Pearson correlation between predicted and observed values in each fold.

mmesCV

mmesCV(dados, G = NULL, W = NULL, CV = NULL, fold.n = 5, looEnv = TRUE, covGE = FALSE)

• dados: phenotypic data frame containing 'gid' (genotype) and 'env' (environment) columns.
• G: genomic additive relationship matrix. If NULL, the identity matrix is used.
• W: environmental similarity matrix. If NULL, the identity matrix is used.
• CV: cross-validation scheme identifier (e.g., '0', '1', '2', '00').
• fold.n: number of folds (partitions) for CV. Default is 5.
• looEnv: for CV = '00' only; indicates whether leave-one-out cross-validation is applied specifically to environments, while genotypes are partitioned separately into folds defined by fold.n.
• covGE: logical, whether to construct GxE covariance structures following Jarquín et al. (2014).

Performs cross-validation for multi-environment data. Supports different CV schemes, including random folds, genotype-based, environment-based, and combined genotype-environment folds. Fits mixed models with additive genomic and environmental covariance structures. Returns accuracy metrics for predictions and GxE effects per fold.

Suggested References / Referências Sugeridas

• Jarquín, D., Crossa, J., Lacaze, X., Du Cheyron, P., Daucourt, J., Lorgeou, J., Piraux, F., Guerreiro, L., Pérez, P., Calus, M., Burgueño, J., de los Campos, G., 2014. A reaction norm model for genomic selection using high-dimensional genomic and environmental data. Theoretical and Applied Genetics 127, 595–607. https://doi.org/10.1007/s00122-013-2243-1

• Costa-Neto, G., Fritsche-Neto, R., Crossa, J., 2021a. Nonlinear kernels, dominance, and envirotyping data increase the accuracy of genome-based prediction in multi-environment trials. Heredity (Edinb) 126, 92–106. https://doi.org/10.1038/s41437-020-00353-1

• Covarrubias-Pazaran, G, 2025. Quantitative genetics using the sommer package. Available at: https://cran.r-project.org/web/packages/sommer/vignettes/sommer.qg.html