Compute genomic prediction different methods
compute_GP_methods.RdCompute genomic prediction different methods
Usage
compute_GP_methods(
geno,
pheno,
traits,
GP.method,
nreps = 10,
nfolds = 10,
h = 1,
nb.mtry = 10,
nIter = 6000,
burnIn = 1000,
ntree = 100,
p2d.temp = NULL,
nb.cores = 1,
p2f.stats = NULL
)Arguments
- geno
genomic data with genotypes in row (GID in rownames) and marker in columns. Values should be column centered and scaled.
- pheno
phenotypic data with genotypes in row (in GID column) and traits in columns. Phenotypic value should be corrected for year and location effects beforehand.
- traits
character vector of trait names
- GP.method
character vector of length one of genomic prediction methods to use. Must be one of "rrBLUP", "GBLUP", "RKHS", "RKHS-KA", "RandomForest", "BayesA", "BayesB" or "LASSO".
- nreps
number of repetitions for cross-validation, default is 10
- nfolds
number of folds for cross-validation, default is 10
- h
bandwith parameter for RKHS, default is 1. If multiple values are provided, method will be RKHS Kernel Averaging.
- nb.mtry
number of mtry for RandomForest, default is 10
- nIter
number of iterations for RKHS, default is 6000
- burnIn
number of burn-in iterations for RKHS, default is 1000
- ntree
number of trees for RandomForest, default is 100
- p2d.temp
path to directory to export temporary genomic prediction results, default is NULL (could cause error in parallelization if NULL).
- nb.cores
number of cores to parallelize the computation, default is 1 (no parallelization)
- p2f.stats
path to file to export genomic prediction results, default is NULL
Value
a list with the following elements: obspred with observed vs. predicted genotypic values, and gp.stats with genomic prediction statistics