GenomicSelection

Forward-simulation breeding programs for genomic selection (GS) in soybean, investigating long-term genetic gain and genetic erosion under different GS models, selection strategies, and selection intensities.

Based on the paper: "Impact of genomic prediction model, selection intensity and breeding strategy on the long-term genetic gain and genetic erosion in soybean breeding."

Installation

# Install from local directory
devtools::install_local('.')

Usage

library(GenomicSelection)

# 1. Build default model presets (Random baseline included; external models added below)
models <- default_models()

# 2. Add external GS models (from bWGR or custom fitting functions)
#    models$GBLUP  <- function(y, gen, ...) bWGR::emML(y, gen, ...)
#    models$BayesA <- function(y, gen, ...) bWGR::emBA(y, gen, ...)

# 3. Configure simulation parameters
params <- list(
    Number_of_runs          = 2L,
    Number_of_generations   = 5L,
    Intensity               = c(0.025, 0.10),     # 2.5%, 10%
    FS                      = list('1' = c(30L, 20L)),
    POPBestIntensity        = 0.3,
    NCgs                    = 3L,
    nCores                  = 4L,                 # local parallel workers
    models                 = models
)

# 4. Run simulation
results <- run_simulation(params)
print(results)

# 5. Analyze results (from saved CSVs)
analyze_results(
    results_dir   = 'output/',
    generations   = c(2L, 3L, 5L),
    output_prefix = 'ALL'
)

Package Functions

Function	Description
run_simulation(params)	Run the forward-simulation breeding program
default_models()	Return default model presets (Random baseline)
analyze_results()	Read CSV outputs, compute metrics, generate plots

Simulation Parameters

Parameter	Default	Description
Number_of_runs	2	Replicated scenarios
Number_of_generations	5	Breeding cycles
Intensity	c(0.025, 0.10)	Selection intensity (proportion)
FS	list('1'=c(30L, 20L))	F1 and F2 family sizes
POPBestIntensity	0.3	Pre-selected family proportion (WPSF)
NCgs	3L	Breeding cycles in estimation set
Ne	106L	Effective population size
segSites	1000L	Segregating sites per chromosome
nInd	200L	Founding individuals
nSnpPerChr	300L	SNPs per chromosome
mean_	60	Trait mean (SoyNAN yield)
var_	77	Trait genetic variance
varGxE	77	GxE interaction variance
varEnv	200	Environmental variance
nCores	NULL (auto)	Local parallel workers (macOS, Linux without a PBS/Lambda cluster, and Windows). Defaults to 2 local workers, or 6 on Windows

Selection Strategies

• AF (Across Family): Select best individuals across all families
• WF (Within Family): Select best individuals within each family
• WPSF (Within Pre-Selected Families): Pre-select top 30% best families, then select within those families

Model Fitting Functions

Model fitting functions are user-supplied. Each takes a phenotype vector y, genotype matrix gen, and optional args, returning list(hat = fitted_values).

Model	Fitting Function	Source
Random	Built-in baseline (noise sample)	Package default
GBLUP	bWGR::emML(...)	External (bWGR)
BayesA	bWGR::emBA(...)	External (bWGR)

Dependencies

• AlphaSimR — Population simulation and breeding operations
• bWGR — Genomic selection model fitting (external dependency)
• ggplot2 — Visualization
• plyr / reshape2 — Data reshaping
• ScottKnott — Statistical clustering tests
• verification — CRPS accuracy metrics

Acknowledgements

• Eder Silva — Primary author
• Alencar Xavier — Co-developer
• Marcos Ventura Farias — Collaborator