In gdlc/BGLR-R: Bayesian Generalized Linear Regression

knitr::opts_chunk$set(echo = TRUE,eval=FALSE,class.source = "numberLines lineAnchors")

Box 1: Multi-trait GBLUP with unstructured and structured covariances

library(BGLR)
data(wheat)
K<-tcrossprod(scale(wheat.X,center=TRUE)) 
K<-K/mean(diag(K))
Y<-wheat.Y # 4 traits

# Fitting a GBLUP un-structured cov-matrices 
 LP<-list(mar=list(K=K,model="RKHS"))
 set.seed(123)
 fmUN<-Multitrait(y=Y,ETA=LP,nIter=10000,burnIn=5000,
                  saveAt="UN_",verbose=FALSE)

# Retrieving estimates and posterior SD
 fmUN$resCov$R          # residual co-variance matrix
 fmUN$resCov$SD.R

 fmUN$ETA$mar$Cov$Omega # genomic covariance matrix
 fmUN$ETA$mar$Cov$SD.Omega

 fmUN$ETA$mar$u         # predicted random effects

Box 2: Multi-trait GBLUP with structured covariance matrices

#(continued from Box 1)

# Genetic (co)variance recursive (not fully), Residual (co)variance diagonal
 # Matrix specifying loading among traits 2=>3,2=>4,3=>4
 M1<-matrix(nrow = 4, ncol = 4, FALSE)
 M1[3,2]<-M1[4,2]<-M1[4,3]<-TRUE
 CovREC<-list(type="REC",M=M1)
 LP<-list(mar=list(K=K,model="RKHS",Cov=CovREC))

 CovDIAG<-list(type="DIAG")

 set.seed(456)
 fmRD<-Multitrait(y=Y,ETA=LP,nIter=10000,burnIn=5000,
                  resCov=CovDIAG,saveAt= "REC_DIAG_",
                  verbose=FALSE)

 fmRD$resCov$R
 fmRD$ETA$mar$Cov$Omega   # genomic covariance
 fmRD$ETA$mar$Cov$W       # recursive genetic effects 
 fmRD$ETA$mar$u           # predicted genetic effects
 fmRD$ETA$mar$Cov$PSI     # scaling factors


# Omega-FA(2), R-diagonal 
 M2<-matrix(nrow=4,ncol=1,FALSE)
 M2[2:4,1]<-TRUE
 CovFA<-list(type="FA",M=M2)
 LP<-list(mar=list(K=K,model="RKHS",Cov=CovFA))

 set.seed(789)
 fmFAD<-Multitrait(y=Y,ETA=LP,nIter=10000,burnIn=5000,
                   resCov=CovDIAG,saveAt= "FA_DIAG_",
                   verbose=FALSE)
 fmFAD$resCov$R
 fmFAD$ETA$mar$Cov$Omega   # genomic covariance
 fmFAD$ETA$mar$Cov$W       # factor scores 
 fmFAD$ETA$mar$u           # predicted genetic effects
 fmFAD$ETA$mar$Cov$PSI     # scaling factors

Box 3: Fitting a Multitrait Spike Slab model

fmSS<-Multitrait(y=Y,ETA=list(list(X=X,model='SpikeSlab',
                 saveEffects=TRUE)),nIter=12000,burnIn=2000)

Box 4: Prediction

K<-tcrossprod(X)
K<-K/mean(diag(K))

LP<-list(list(K=K,model="RKHS"))

#Fit multivariate GBLUP with UN-structured covariance matrixes
fmG<-Multitrait(y=YNa,ETA=LP,nIter=10000,burnIn=5000,thin=10,
                verbose=FALSE)

#Missing values for trait 3
whichNa3<-fmG$missing_records[fmG$patterns[,3]]
Y[whichNa3,3]          #Observed values
fmG$ETAHat[whichNa3,3] #Predicted values