K-Fold Cross-Validation for Haplotype-Based Genomic Prediction

Estimates the predictive ability of the haplotype GBLUP model via k-fold cross-validation. In each fold, a subset of individuals is masked from the phenotype and predicted from the haplotype GRM; Pearson correlation between predicted and observed BLUEs is returned as the predictive ability (PA). Runs per trait when multiple traits are supplied.

Usage

cv_haplotype_prediction(
  geno_matrix,
  snp_info,
  blocks,
  blues,
  k = 5L,
  n_rep = 1L,
  top_n = NULL,
  min_freq = 0.05,
  min_snps = 3L,
  id_col = "id",
  blue_col = "blue",
  blue_cols = NULL,
  seed = 42L,
  verbose = TRUE
)

Arguments

geno_matrix

Numeric matrix (individuals x SNPs), MAF-filtered dosage.

snp_info

Data frame with columns SNP, CHR, POS.

blocks

LD block table from run_Big_LD_all_chr.

blues

Pre-adjusted phenotype means. Accepts the same four formats as run_haplotype_prediction: named numeric vector, single-trait data frame, multi-trait data frame, or named list.

k

Integer. Number of folds. Default 5L.

n_rep

Integer. Number of CV replications (each with a different random fold assignment). Default 1L.

top_n

Integer or NULL. Maximum haplotype alleles per block passed to build_haplotype_feature_matrix. Default NULL (all alleles above min_freq).

min_freq

Numeric. Minimum haplotype allele frequency. Default 0.05.

min_snps

Integer. Minimum SNPs per block for haplotype extraction. Default 3L.

id_col

Character. Name of the individual ID column when blues is a data frame. Default "id".

blue_col

Character. Name of the BLUE column for single-trait data frames. Default "blue".

blue_cols

Character vector. Trait column names for multi-trait data frames. Default NULL (auto-detect all numeric non-ID columns).

seed

Integer. RNG seed for reproducible fold assignment. Default 42L.

verbose

Logical. Print progress. Default TRUE.

Value

A named list of class LDxBlocks_cv:

pa_summary

Data frame: trait, rep, fold, n_train, n_test, PA (Pearson r), RMSE.

pa_mean

Data frame: mean PA and RMSE per trait across all folds and replications.

gebv_all

Data frame of out-of-fold GEBVs for all individuals and traits (one row per individual x trait).

k

Number of folds used.

n_rep

Number of replications.

See also

run_haplotype_prediction, build_haplotype_feature_matrix

Examples

# \donttest{
data(ldx_geno, ldx_snp_info, ldx_blocks, ldx_blues, package = "LDxBlocks")
cv <- cv_haplotype_prediction(
  geno_matrix = ldx_geno,
  snp_info    = ldx_snp_info,
  blocks      = ldx_blocks,
  blues       = ldx_blues,
  k           = 5L,
  id_col      = "id",
  verbose     = FALSE
)
cv$pa_mean
#>   trait        PA      RMSE     PA_sd    RMSE_sd
#> 1   RES 0.3208779 0.9430814 0.1538289 0.07182249
#> 2   YLD 0.1742259 1.0030557 0.1953338 0.09088871
# }