blupga_EFF: BLUP|GA using top SNPS according to estimated SNP effect

In dkainer/BLUPGA: BLUP|GA genomic prediction model

Description

This function runs the BLUP|GA method where SNPs with the greatest squared effect size are weighted in the GRM prior to GBLUP.

Usage

blupga_EFF(G, phenodata, valset = NULL, genomat, bsq, perc, flank = TRUE,
  verbose = TRUE)

Arguments

G	G-matrix constructed using all available SNPs and samples Defaults to NULL.
phenodata	data frame with 2 columns. One col must be named 'ID' and contain sample IDs. Another column must be named 'y' and contain the phenotypes. Defaults to NULL.
valset	vector of indices that defines which rows in phenodata will be set to NA and used for cross-validation Defaults to NULL.
genomat	matrix of genotypes in -1,0,1 format (i.e. 0 is heterozygous, 1 and -1 are opposing homozygous states).
bsq	vector of squared marker effects the same length as the number of SNPs in genomat. Can be obtained with est_SNPeffects().
perc	proportion of SNPs to be weighted (between 0 and 1), where 0.05 means the top 5 percent of SNPs will be weighted. Defaults to NULL.
flank	choose to include the immediate SNPs to the left and right of a top 'perc' SNP. Defaults to TRUE.
verbose	just leave this for now!

Value

A data frame containing the correlation between the predicted phenotype and the true phenotype of the individuals in the valset.

Since BLUP|GA is run for each value of omega (W) from 0.0 to 1.0 in increments of 0.10, each row of the returned data frame contains the cross-validation correlation at one value of omega (W). This allows the user to find the value of W at which the predictive ability (COR) is maximised.

W

omega weighting for selected SNPS in candidate genes (0.0–1.0)

COR

cross validation predictive ability (0.0–1.0)

Examples

# get example genotype matrix and phenotype data
data(M)
data(pheno)


# choose a validation set of 20 random individuals
val <- sample(1:nrow(pheno), 20)

# Run a GWAS to estimate squared SNP effects for all SNPs.
# By calling est_SNPeffects() this process will be done using the individuals NOT in your 'val' set.
# Of course you can use your own method to generate the full vector of SNP effects using a training set of individuals.
bsq <- est_SNPeffects(pheno, M, val)

# make a standard GRM for all SNPs
G <- make_GRM(M)

# run BLUPGA where the top 1 percent of SNPs (according to effect size) are weighted
results <- blupga_EFF(G, pheno, val, M, bsq, 0.01, flank=FALSE)

# run BLUPGA where the top 0.1 percent of SNPs (according to effect size) are weighted and
# SNPs immediately upstream and downstream of those top SNPs are also weighted.
results <- blupga_EFF(G, pheno, val, M, bsq, 0.001, flank=TRUE)

dkainer/BLUPGA documentation built on Jan. 3, 2020, 1:09 a.m.