GeneABEL Exercise
From Statistical Genetics Courses
GeneABEL Exercise
R:
# Load files
library(GenABEL)
convert.snp.tped(tped = "gwa_gabel_qtl.tped", tfam = "gwa_gabel_qtl.tfam", out = "gwa_gabel_qtl.raw", strand = "u")
g.dat <- load.gwaa.data(phen = "gwa_gabel_qtl.praw", gen = "gwa_gabel_qtl.raw", force = T)
slotNames(g.dat)
slotNames(g.dat@gtdata)
colnames(g.dat@phdata)
# sample size
sample.size <- g.dat@gtdata@nids
# number of SNPs
snps.total <- g.dat@gtdata@nsnps
print(c(sample.size, snps.total))
# Trait
summary(g.dat@phdata$disease)
hist(g.dat@phdata$disease, main="Quantitative Phenotype data summary", xlab = "Systolic pressure measure", freq = F,breaks=20, col="gray")
rug(g.dat@phdata$disease)
###
# tests for association
###
# GLM test
test.snp <- scan.glm('disease ~ CRSNP', family = gaussian(), data = g.dat)
names(test.snp)
alpha <- 5e-8
test.snp$snpnames[test.snp$P1df < alpha]
test.snp$P1df[test.snp$P1df < alpha]
# Score test
test.qt <- qtscore(disease, data = g.dat, trait = "gaussian")
slotNames(test.qt)
names(test.qt@results)
test.qt@lambda
descriptives.scan(test.qt)
rownames(results(test.qt))[results(test.qt)$P1df < alpha]
results(test.qt)$P1df[results(test.qt)$P1df < alpha]
results(test.qt)$Pc1df[results(test.qt)$Pc1df < alpha]
# QQ plot
obs <- sort(results(test.qt)$P1df)
ept <- c(1:length(obs)) / (length(obs) + 1)
plot(-log10(ept), -log10(obs), main = "GWAS QQ plot, qtl", xlab="Expected -log10(pvalue)", ylab="Observed -log10(pvalue)")
abline(0, 1, col = "red")
abline(h = 8, lty = 2)
# Manhattan plot
plot(test.qt, col = "black")
# Adding confounders
test.qt.sex <- qtscore(disease ~ sex, data = g.dat, trait = "gaussian")
rownames(results(test.qt.sex))[results(test.qt)$P1df < alpha]
summary(lm(disease ~ sex, data = g.dat))
###
# MDS
###
gkin <- ibs(g.dat, weight = "freq")
gkin[1:10,1:10]
cps.full <- cmdscale(as.dist(.5 - gkin), eig = T, k = 10)
names(cps.full)
cps <- cps.full$points
plot(cps[,1], cps[,2], pch = g.dat@phdata$popn)
legend(-0.16, 0.06, c("TSI","MEX", "CEU"), pch = c(1,2,3))
###
# Corrected test
###
# Incorporating PCs as predictors
colnames(cps)<-c('C1','C2','C3','C4','C5','C6','C7','C8','C9','C10')
gpc.dat <- g.dat
gpc.dat@phdata<-cbind(g.dat@phdata, cps)
test.pc.a <- scan.glm('disease ~ CRSNP + C1 + C2 + C3 + C4 + C5', family=gaussian(), data = gpc.dat)
test.pc.a$snpnames[test.pc.a$P1df < alpha]
test.pc.a$P1df[test.pc.a$P1df < alpha]
test.pc.b <- qtscore(disease ~ C1 + C2 + C3 + C4 + C5, data = gpc.dat, trait = "gaussian")
test.pc.b@lambda
# scree plot
plot(cps.full$eig[1:10]/sum(cps.full$eig), axes = F, type = "b", xlab = "Components", ylim = c(0,0.05), ylab = "Proportion of Variations", main = "MDS analysis scree plot")
axis(1, 1:10)
axis(2)
# cumulative plot
plot(cumsum(cps.full$eig[1:10])/sum(cps.full$eig), axes = F, type = "b", ylim = c(0,0.2), xlab = "Components", ylab = "Proportion of Variations", main = "MDS analysis cumulative plot")
axis(1, 1:10)
axis(2)
# Genomic control
# Uncorrected GIF
test.qt@lambda
# Corrected p-value
row.names(results(test.qt))[results(test.qt)$Pc1df < alpha]
results(test.qt)$Pc1df[results(test.qt)$Pc1df < alpha]
# Check for inflation of statistic
obs <- sort(results(test.qt)$chi2.1df)
ept <- sort(qchisq(1:length(obs) / (length(obs) + 1), df = 1))
plot(ept, obs, main = "Genomic control (slope is the inflation factor)", xlab="Expected chisq, 1df", ylab="Observed chisq, 1df")
abline(0, 1, col = "red")
abline(0, test.qt@lambda[1], lty = 2)
# Definition of GIF
# Conventional definition
median(results(test.qt)$chi2.1df)/0.456
# GenABEL definition
lm(obs~ept)$coef[2]
# QQ plot
obs <- sort(results(test.qt)$Pc1df)
ept <- c(1:length(obs)) / (length(obs) + 1)
plot(-log10(ept), -log10(obs), main = "GWAS QQ plot adj. via Genomic Control", xlab="Expected -log10(pvalue)", ylab="Observed -log10(pvalue)")
abline(0, 1, col = "red")
abline(h = 8, lty = 2)
# EIGENSTRAT
adj.gkin = gkin
diag(adj.gkin) = hom(g.dat)$Var
# naxes = 3 is default value
test.eg <- egscore(disease, data = g.dat, kin = adj.gkin, naxes = 2)
descriptives.scan(test.eg)
snp.eg <- row.names(results(test.eg))[results(test.eg)$P1df < alpha]
pvalue.eg <- results(test.eg)$P1df[results(test.eg)$P1df < alpha]
lambda.eg <- test.eg@lambda
snp.eg
pvalue.eg
lambda.eg
# Change #PCs
for (k in 1:10){
test.tmp <- egscore(disease, data = g.dat, kin = adj.gkin, naxes = k)
print(test.tmp@lambda$estimate)
}
# QQ plot
obs <- sort(results(test.eg)$Pc1df)
ept <- c(1:length(obs)) / (length(obs) + 1)
qqplot(-log10(ept), -log10(obs), main = "GWAS QQ plot adj. w/ EIGENSTRAT", xlab="Expected -log10(pvalue)", ylab="Observed -log10(pvalue)")
abline(0, 1, col = "red")
abline(h = 8, lty = 2)
# Manhattan plot comparison
plot(test.qt, col = "black")
add.plot(test.eg, col = "gray", pch = 3)
legend("topright", c("Original plot","After correction w/ EIGENSTRAT"), pch = c(1,3))
###
# Basic test, binary trait
###
# load files to GenABEL
convert.snp.tped(tped = "gwa_gabel.tped", tfam = "gwa_gabel.tfam", out = "gwa_gabel.raw", strand = "u")
b.dat <- load.gwaa.data(phen = "gwa_gabel.praw", gen = "gwa_gabel.raw", force = T)
slotNames(b.dat)
slotNames(b.dat@gtdata)
colnames(b.dat@phdata)
# sample size
b.dat@gtdata@nids
# number of cases and controls
case.size <- length(which(b.dat@phdata$disease == 1))
control.size <- length(which(b.dat@phdata$disease == 0))
case.size
control.size
# number of SNPs
snpsb.total <- b.dat@gtdata@nsnps
# GLM test
testb.snp <- scan.glm('disease ~ CRSNP', family = binomial(), data = b.dat)
names(testb.snp)
alpha <- 5e-8
testb.snp$snpnames[testb.snp$P1df < alpha]
testb.snp$P1df[testb.snp$P1df < alpha]
# Score test
testb.qt <- qtscore(disease, data = b.dat, trait = "binomial")
slotNames(testb.qt)
descriptives.scan(testb.qt)
row.names(results(testb.qt))[results(testb.qt)$P1df < alpha]
results(testb.qt)$P1df[results(testb.qt)$P1df < alpha]
results(testb.qt)$Pc1df[results(testb.qt)$Pc1df < alpha]
