Stage2: Stage 2 analysis of multi-environment trials

In jendelman/StageWise: Two-stage analysis of multi-environment trials for genomic selection and GWAS

Stage 2 analysis of multi-environment trials

Description

Stage 2 analysis of multi-environment trials

Usage

Stage2(
  data,
  vcov = NULL,
  geno = NULL,
  fix.eff.marker = NULL,
  silent = TRUE,
  workspace = "500mb",
  non.add = "g.resid",
  max.iter = 20,
  covariates = NULL,
  pairwise = FALSE
)

Arguments

data	data frame of BLUEs from Stage 1 (see Details)
vcov	named list of variance-covariance matrices for the BLUEs
geno	output from read_geno
fix.eff.marker	markers in geno to include as additive fixed effect covariates
silent	TRUE/FALSE, whether to suppress ASReml-R output
workspace	Memory limit for ASRreml-R variance estimation
non.add	one of the following: "none","g.resid","dom"
max.iter	maximum number of iterations for asreml
covariates	names of other covariates in data
pairwise	TRUE/FALSE should multi-trait analysis proceed pairwise

Details

Stage 2 of the two-stage approach described by Damesa et al. 2017, using ASReml-R for variance component estimation. The variable data has three mandatory column: id, env, BLUE. Optionally, data can have a column labeled "loc", which changes the main effect for genotype into a separable genotype-within-location effect, using a FA2 covariance model for the locations. Optionally, data can have a column labeled "trait", which uses an unstructured covariance model. The multi-location and multi-trait analyses cannot be combined. Missing data are allowed in the multi-trait but not the single-trait analysis. The argument geno is used to partition genetic values into additive and non-additive components. Any individuals in data that are not present in geno are discarded.

The argument vcov is used to partition the macro- and micro-environmental variation, which are called GxE and residual in the output. vcov is a named list of variance-covariance matrices for the BLUEs within each environment, with id for rownames (single trait) or id:trait. The order in vcov and data should match. Both data and vcov can be created using the function Stage1.

Because ASReml-R can only use relationship matrices defined in the global environment, this function creates and then removes global variables when either vcov or geno is used. By default, the workspace memory for ASReml-R is set at 500mb. If you get an error about insufficient memory, try increasing it. ASReml-R version 4.1.0.148 or later is required.

The covariates option is only available for single trait/loc analysis.

Argument pairwise was added in package version 1.04, which specifies that multi-trait analysis is performed as multiple bivariate analyses, which often converges better. The returned object is a list of the results from the bivariate analyses, as well as "vars" for all traits, which is needed for blup_prep.

Value

List containing

aic

AIC

vars

variance components for blup_prep, as variable of class class_var

params

Estimates and SE for fixed effects and variance components

random

Random effect predictions

loadings

scaled loadings for the FA2 multi-loc model

References

Damesa et al. 2017. Agronomy Journal 109: 845-857. doi:10.2134/agronj2016.07.0395

jendelman/StageWise documentation built on Feb. 23, 2025, 11 a.m.