BLUP|GA

R implementation of BLUP|GA genomic prediction.

installation

Can be installed directly from github with Bioconductor install manager. Package 'cpgen' needs to be installed first.

BiocManager::install("cheuerde/cpgen")
BiocManager::install("dkainer/BLUPGA") 

citing this package

Kainer, D., Stone, E. A., Padovan, A., Foley, W. J., & Külheim, C. (2018). Accuracy of Genomic Prediction for Foliar Terpene Traits in Eucalyptus polybractea. G3: Genes, Genomes, Genetics, 8(8), 2573-2583.

what is BLUP|GA?

BLUP|GA means Best Linear Unbiased Prediction given Genetic Architecture. It is a genomic prediction model that is basically a weighted version of GBLUP. The original BLUP|GA publication can be found here:

Accuracy of Whole-Genome Prediction Using a Genetic Architecture-Enhanced Variance-Covariance Matrix Zhang et al (2015) [https://doi.org/10.1534/g3.114.016261]

In a nutshell, GBLUP uses a genomic relationship matrix (GRM) to describe the covariance between individuals in the study population. All SNPs used in calculating the GRM are treated the same as each other. With BLUP|GA the calculation of the GRM is done with two separate sub-GRMs:

• S-matrix: this is a GRM made with selected SNPs that have an (assumed) additive effect on the trait. These are the genetic architecture SNPs. They are given weights according to a metric such as their squared effect size from a GWAS model.
• G-matrix: this is a GRM made with all SNPs. These are the background SNPs that probably don't affect the trait.

A further weighting factor (w) is applied to S so that the genetic architecture SNPs can be given more or less importance relative to the background SNPs in G. The S and G matrices are then recombined into the final GRM (called the T-matrix):

T = w(S) + (1-w)(G)

the effect of w

w ranges from 0 to 1.

• When w=0 the selected SNPs in the S matrix are given no overall weight and the model is simply GBLUP
• When w=1 the selected SNPs in the S matrix are given full weight and the background SNPs are ignored.

Using this R package

The basic method is blupga_EFF().

To run BLUP|GA you need:

• a genotype matrix (in -1,0,1} format.
• a phenotype data frame with the first column containing your sample IDs..
• a vector of SNP effects the same length as the number of SNPs in the genotype matrix. you can generate these with est_SNPeffects() or bring your own from other methods

The method blupga_EFF() lets you specify what percent of your SNPs you want to be used in the S matrix. e.g. the top 1% of SNPs when ranked by squared effect size.