Please note, we no longer recommend using MultiBLUP, therefore this page exists only for completeness.

The most basic use of MultiBLUP is when we have already computed a number of kinship matrices (which because the aim is prediction we advise doing with weightings turned off) and wish to perform generalised BLUP. First we use REML to estimate the variance components.

--reml <outfile> --pheno <phenofile> --mgrm <grmlist>

where the file <kinstems> provides the stems of the pre-computed kinship matrices and <phenofile> provides the phenotypic values (in PLINK format). The variance estimates will be saved in <outfile>.reml, with the random effects in <outfile>.indi.blp.

Use the option --covar <covarfile> to provide covariates (in PLINK format). Use --keep <keepfile> and/or --remove <removefile> to consider only a subset of samples; this is useful if wishing to fit the model on a training sample and then measure predictive accuracy for a test sample.

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Next we use BLUP to obtain estimates of predictor effect sizes.

--calc-blups <outfile> --mgrm <kinstems> --remlfile <outfile>.reml --bfile test

This step requires one of –bfile/–chiamo/–sp/–speed <datastem> , to provide the genetic data files from which the kinship matrices were calculated (see File Formats).

The SNP effect sizes will be saved in <outfile>.blup, with the random effects in <outfile>.pred.

If <kinstems> provides the stem for only one kinship matrix (or if instead of --mgrm <kinstems> you use --grm <kinfile>), then the above commands will be equivalent to running (standard) BLUP.

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

**Example:**

Here we use the binary PLINK files human.bed, human.bim and human.fam, and the phenotype quant.pheno from the Test Datasets. We also use the kinship matrices with stems chr21 and chr22 created in the example for Genomic Partitioning.

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

First we perform REML, regressing the phenotype quant.pheno on the kinship matrices with stems chr21 and chr22

echo "chr21

chr22" > mlist.txt

./ldak.out --reml reml5 --pheno quant.pheno --mgrm mlist.txt

Next we estimate effect sizes using the command

./ldak.out --calc-blups reml5 --mgrm mlist.txt --remlfile reml5.reml --bfile human

We have now performed 2-way MultiBLUP. The final effect size estimates are saved in reml5.blup (the file reml5.blup.full divides these effects into the contribution from each kinship matrix).