-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathModels_of_prediction_body_weight.R
182 lines (139 loc) · 6.93 KB
/
Models_of_prediction_body_weight.R
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
setwd('C:/Users/MariaIzabel/Desktop/MASTER/QUANTITATIVE GENOMICS/Genomic_Prediction_Models/')
dat = read.table("http://www.bayz.biz/mouse_allgeno.txt", header = T)
dat <- dat[!is.na(dat$FI),]
dat_train = dat[dat$CV1 == 1, ]
dat_test = dat[dat$CV1 == 2, ]
geno_train = dat_train[, 9:1821]
geno_test = dat_test[, 9:1821]
map = read.table("http://www.bayz.biz/mousesnps.map",header=F)
yteste <- dat_test$BW # Selecting the weigth as phenotypic measure for the test set
ytreino <- dat_train$BW # Selecting the weigth as phenotypic measure for the train set
#replacing zeros by the mean
yteste[is.na(yteste)] <- mean(yteste, na.rm = TRUE)
ytreino[is.na(ytreino)] <- mean(ytreino, na.rm = TRUE)
#Estimating the heritability of the trait based on markers
require(heritability)
geno.vector = seq(1, nrow(dat), by = 1)
# Calculating the Genetic Relatedness Matrix (G or K):
genotypes = dat[,9:1821]
geno_centered = scale(genotypes,scale=F, center=T) # Z
# Inputting zero in the missing genotypes
geno_centered[is.na(geno_centered)] <- 0 # Z correct for missing values
mean(geno_centered) #almost zero
Z = geno_centered %*% t(geno_centered)
p = as.vector(colMeans(genotypes,na.rm=T)/2)
q = 1- p
var = sum(2*p*q)
K = Z/var
row.names(K) = geno.vector
colnames(K) = geno.vector
weight_heritability = marker_h2(dat$BW, geno.vector, covariates = dat$Sex + dat$batch, K, alpha = 0.05,
eps = 1e-06, max.iter = 100, fix.h2 = FALSE, h2 = 0.5)
weight_heritability$h2
FI_heritability = marker_h2(dat$FI, geno.vector, covariates = dat$Sex + dat$batch, K, alpha = 0.05,
eps = 1e-06, max.iter = 100, fix.h2 = FALSE, h2 = 0.5)
FI_heritability$h2
# The real variance of the data is:
#The real variance of the data
var(dat$BW, na.rm= T)
require(BGLR)
#install.packages('rrBLUP')
require(rrBLUP) # To be able to use the mixed.solve function
nIter<-20000 ; burnIn<-2000
# It makes a 'geno_noNA' table that can be used in BGLR.
geno_train_noNA = t(geno_train) - colMeans(geno_train, na.rm = T)
geno_train_noNA[is.na(geno_train_noNA)] = 0
geno_train_noNA = t(geno_train_noNA)
# also make "noNA" version of test genotypes; it avoids that predicted GEBV fall out
# due to missing genotypes.
geno_test_noNA = t(geno_test) - colMeans(geno_test, na.rm = T)
geno_test_noNA[is.na(geno_test_noNA)] = 0
geno_test_noNA = t(geno_test_noNA)
fixedmod = model.matrix(~factor(dat_train$Sex) + factor(dat_train$batch))
print(paste('Modelo BayesB'))
ETA = list(list(X = fixedmod, model = "FIXED"), list(X = geno_train_noNA, model='BayesB'))
fmRBB <- BGLR(y=ytreino,ETA=ETA, nIter=nIter, burnIn=burnIn,verbose= FALSE, saveAt="BB")
print(paste('Modelo BayesA'))
ETA[[1]]$model<-'BayesA'
fmRBA <- BGLR(y=ytreino,ETA=ETA, nIter=nIter, burnIn=burnIn,verbose= FALSE, saveAt="BA")
print(paste('Modelo BRR'))
ETA[[1]]$model<-'BRR'
fmRBRR <- BGLR(y=ytreino,ETA=ETA, nIter=nIter, burnIn=burnIn,verbose= FALSE, saveAt="BRR")
print(paste('Modelo BayesC'))
ETA[[1]]$model<-'BayesC'
fmRBC <- BGLR(y=ytreino,ETA=ETA, nIter=nIter, burnIn=burnIn,verbose= FALSE, saveAt="BC")
print(paste('Modelo BayesLASSO'))
ETA[[1]]$model<-'BL'
fmRBL <- BGLR(y=ytreino,ETA=ETA, nIter=nIter, burnIn=burnIn, verbose= FALSE, saveAt="BL")
print(paste('Modelo RRBlup'))
rrblup = mixed.solve(ytreino, Z=geno_train_noNA)
assign(paste('fmRR'), rrblup)
save.image("Weight.RData")
load("Weight.RData")
###########################################################
# PARTE 5: OBtaining GEBVs
############################################################
GEBVBB <-geno_test_noNA%*%fmRBB$ETA[[2]]$b
GEBVBA <-geno_test_noNA%*%fmRBA$ETA[[2]]$b
GEBVBRR <-geno_test_noNA%*%fmRBRR$ETA[[2]]$b
GEBVBC <-geno_test_noNA%*%fmRBC$ETA[[2]]$b
GEBVBL <-geno_test_noNA%*%fmRBL$ETA[[2]]$b
GEBVRRBlup <- geno_test_noNA%*%rrblup$u #GEBV for individuals of test population
GEBVRRBlup2 <- geno_train_noNA%*%rrblup$u #GEBV for individuals of train population
############################################################
# PARTE 6: Obtaining Correlations test set
############################################################
models_breeding_values = cbind(GEBVBB, GEBVBA, GEBVBRR, GEBVBC, GEBVBL, GEBVRRBlup)
colnames(models_breeding_values) = c('GEBVBB', 'GEBVBA', 'GEBVBRR', 'GEBVBC', 'GEBVBL', 'GEBVRRBlup')
models_breeding_values = as.data.frame(models_breeding_values)
cor.tst = vector()
for(i in 1:6){
cor.tst[i] <- cor(models_breeding_values[,i], yteste)
}
cor.tst
seq_h2 = rep(weight_heritability$h2, 6)
test_cor = rbind(cor.tst, seq_h2)
test_cor = as.data.frame(t(test_cor))
test_cor$Model = c('Bayes B', 'Bayes A', 'Bayes RR', 'Bayes C', 'Bayes Lasso', 'RRBlup')
p1 <- ggplot(test_cor, aes(x = seq_h2, y = cor.tst))
p1 + geom_point(aes(color=factor(Model)), size = 4) + labs(x="Heritability",y="Correlation with test data") + geom_point(colour = "grey90", size = 1.5) + ggtitle("Weight")
# +geom_text(aes(label=cor.tst),hjust=0, vjust=0)
############################################################
# PARTE 7: Obtaining Correlations train set
############################################################
predicted_Y = cbind(fmRBB$yHat, fmRBA$yHat, fmRBRR$yHat, fmRBC$yHat, fmRBL$yHat, GEBVRRBlup2)
colnames(predicted_Y) = c('GEBVBB', 'GEBVBA', 'GEBVBRR', 'GEBVBC', 'GEBVBL', 'GEBVRRBlup')
cor.trn = vector()
for(i in 1:6){
cor.trn[i] <- cor(predicted_Y[,i], ytreino) ## this correlation indicates overfitting
}
train_cor = rbind(cor.trn, seq_h2)
train_cor = as.data.frame(t(train_cor))
train_cor$Model = c('Bayes B', 'Bayes A', 'Bayes RR', 'Bayes C', 'Bayes Lasso', 'RRBlup')
p2 <- ggplot(train_cor, aes(x = seq_h2, y = cor.trn))
p2 + geom_point(aes(color=factor(Model)), size = 4) + geom_point(colour = "grey90", size = 1.5) + ggtitle("Weight") +labs(x="Heritability",y="Correlation with train data")
###############################################
par(mfrow=c(2,3))
plot(fmRBB$yHat, ytreino, xlab="Weight",
ylab="Estimated phenotype", cex=.8, bty="L", main= 'Bayes B')
plot(fmRBA$yHat, ytreino, xlab="Weight",
ylab="Estimated phenotype", cex=.8, bty="L", main= 'Bayes A')
plot(fmRBRR$yHat, ytreino, xlab="Weight",
ylab="Estimated phenotype", cex=.8, bty="L", main= 'Bayes RR')
plot(fmRBC$yHat, ytreino, xlab="Weight",
ylab="Estimated phenotype", cex=.8, bty="L", main= 'Bayes C')
plot(fmRBL$yHat, ytreino, xlab="Weight",
ylab="Estimated phenotype", cex=.8, bty="L", main= 'Bayes Lasso')
plot(GEBVRRBlup2%*%geno_test_noNA, ytreino, xlab="Weight",
ylab="Estimated phenotype", cex=.8, bty="L", main= 'RRBlup')
###############################################
# Gráficos BGLR #
###############################################
bhat <- fmRBB$ETA[[2]]$b
SD.bhat <- fmRBB$ETA[[2]]$SD.b
plot(bhat^2, ylab='Estimated Squared-Marker Effect', type='o', cex=.5, col=4, main='Marker Effects')
tmp <- range(c(ytreino, fmRBB$yHat))
plot(GEBVBB, yteste, ylab='Phenotype', xlab='Predicted Genomic Value')
###################################
# Estimating Genomic prediction usinf FI
######################################