#' Calculate selection statistics (LD) and perform exploratory analyses
#' for two sets of variants via R snpStats package
#' https://bioconductor.org/packages/release/bioc/manuals/snpStats/man/snpStats.pdf
#' @param plinkF (char) path to file with SNP genotype data (PLINK format)
#' @param highSnpDir (char) path to files with pathway SNP lists
#' @param makePlots (logical) set to TRUE to generate plots
#' @return
#' @export
calculateAssoc <- function(plinkF, highSnpDir, makePlots=FALSE) {
# Read PLINK files for high confidence pathway
top_bed <- list.files(path=highSnpDir, pattern="*.bed", full.names=T)
top_bim <- list.files(path=highSnpDir, pattern="*.bim", full.names=T)
top_fam <- list.files(path=highSnpDir, pattern="*.fam", full.names=T)
top_list <- c()
top_r2_matrix <- list()
top_dprime_matrix <- list()
top_pairwise_df <- list()
top_diff_chr <- list()
message("\n-------HIGH CONFIDENCE PATHWAY SNPS-------\n")
for (i in 1:length(top_bed)) {
# Convert PLINK files to snpStats input format
# Output object is a list with 3 elements ($genotypes, $fam, $map)
# NOTE: order is important!
top_list[[i]] <- read.plink(top_bed[i], top_bim[i], top_fam[i])
# Calculate linkage disequilbrium statistics (R squared)
# NOTE: argument 'depth' specifies the max. separation b/w pairs of SNPs
# to be considered, so that depth=1 would specify calculation of LD b/w
# immediately adjacent SNPs. For our purposes we want to determine LD
# b/w all SNPs in each pathway despite their distance from each other,
# so we specify depth as ((number of SNPs)-1)
cat(sprintf("Calculating LD statistics for SNPs in %s pathway...",
basename(file_path_sans_ext(top_bed[i]))))
top_ld_calc[[i]] <- ld(top_list[[i]]$genotypes,
stats=c("D.prime", "R.squared"),
depth=ncol(top_list[[i]]$genotypes)-1)
top_r2_matrix[[i]] <- top_ld_calc[[i]]$R.squared
top_dprime_matrix[[i]] <- top_ld_calc[[i]]$D.prime
cat(" done.\n")
# Create dataframe containing pairwise distance calculations for each
# LD SNP pair
snp_map <- top_list[[i]]$map
# Turn each LD matrix into a data frame
top_r2 <- as.matrix(top_r2_matrix[[i]]) #convert sparseMatrix to regular matrix
top_r2 <- subset(melt(top_r2), value!=0) #for all non-zero values
colnames(top_r2)[3] <- "R2"
top_dprime <- as.matrix(top_dprime_matrix[[i]])
top_dprime <- subset(melt(top_dprime), value!=0)
colnames(top_dprime)[3] <- "Dprime"
# Combine R2 and Dprime stats for each SNP-SNP pair
top_stats <- merge(top_r2, top_dprime, by=c("Var1", "Var2"))
# Generate pariwise distance table for each SNP-SNP pair
colnames(top_stats)[1] <- "snp.name"
snp_map <- subset(snp_map, select=c("snp.name", "chromosome", "position"))
top_pairwise <- merge(snp_map, top_stats, by="snp.name")
colnames(top_pairwise)[1:4] <- c("snp_1", "chr_1", "pos_1", "snp.name")
top_pairwise <- merge(snp_map, top_pairwise, by="snp.name")
colnames(top_pairwise) <- c("snp_1", "chr_1", "pos_1", "snp_2",
"chr_2", "pos_2", "R2", "Dprime")
top_pairwise$dist <- abs(top_pairwise$pos_1 - top_pairwise$pos_2)
top_pairwise_df[[i]] <- top_pairwise %>% mutate(R2 = round(R2, 3))
top_diff_chr[[i]] <- filter(top_pairwise_df[[i]], chr_1 != chr_2) %>%
dplyr::select(R2) %>% unlist
}; cat(" done.\n")
all_top <- do.call("rbind", top_pairwise_df)
#get sample size per pathway
top_diff_num <- sapply(top_diff_chr, length)
#get mean r2 value per pathway
top_diff_mean <- sapply(top_diff_chr, mean)
cat(sprintf("Calculated LD for %i total SNP pairs.\n", nrow(all_top)))
cat(sprintf("%i SNP-SNP interactions in high-conf pathway %i (mean = %g)\n",
top_diff_num, seq(top_bed), top_diff_mean))
cat(sprintf("%i total interchromosomal SNP-SNP pairs.\n", sum(top_diff_num)))
#remove original data objects to clear memory
rm(top_list, top_ld_matrix)
#============================================================================#
# Permute random samples from original PLINK genotype data and calculate LD
message("\n-------RANDOMLY SELECTED SNPS-------\n")
# Large vector, time intensive
start.time <- Sys.time()
test <- read.plink(plinkF)
end.time <- Sys.time()
time.taken <- end.time - start.time
print(time.taken)
rep.num <- 50L #how many permutations to run
sample.num <- 40L #number of SNPs to pick for each permutation
# Generating LD r2 stats for 500 permutations of 100 SNPs each
# later will plot mean null r2 distribution via ggplot
#null <- replicate(rep.num, {
# shuffle <- ld(test$genotypes[, sample(ncol(test$genotypes),
# sample.num, replace=F)],
# stats="R.squared",
# depth=sample.num-1)
#})
null <- list()
null_r2_matrix <- list()
null_dprime_matrix <- list()
null_pairwise_df <- list()
null_diff_chr <- list()
Sys.sleep(5)
for (i in 1:rep.num) {
cat(sprintf("Calculating LD within random sample matrix %i...", i))
null[[i]] <- ld(test$genotypes[, sample(ncol(test$genotypes),
sample.num, replace=F)],
stats=c("D.prime", "R.squared"),
depth=sample.num-1)
null_r2_matrix[[i]] <- null[[i]]$R.squared
null_dprime_matrix[[i]] <- null[[i]]$D.prime
# Create dataframe containing pairwise distance calculations for each
# LD SNP pair
snp_map <- test$map
# Turn each LD matrix into a data frame
null_r2 <- as.matrix(null_r2_matrix[[i]]) #convert sparseMatrix to regular matrix
null_r2 <- subset(melt(null_r2), value!=0) #melt df and remove '0's
colnames(null_r2)[3] <- "R2"
null_dprime <- as.matrix(null_dprime_matrix[[i]])
null_dprime <- subset(melt(null_dprime), value!=0)
colnames(null_dprime)[3] <- "Dprime"
# Combine R2 and Dprime stats for each SNP-SNP pair
null_stats <- merge(null_r2, null_dprime, by=c("Var1", "Var2"))
# Generate pariwise distance table for each SNP-SNP pair
colnames(null_stats)[1] <- "snp.name"
snp_map <- subset(snp_map, select=c("snp.name", "chromosome", "position"))
null_pairwise <- merge(snp_map, null_stats, by="snp.name")
colnames(null_pairwise)[1:4] <- c("snp_1", "chr_1", "pos_1", "snp.name")
null_pairwise <- merge(snp_map, null_pairwise, by="snp.name")
colnames(null_pairwise) <- c("snp_1", "chr_1", "pos_1", "snp_2",
"chr_2", "pos_2", "R2", "Dprime")
# Calculate distance between SNP pairs
null_pairwise$dist <- abs(null_pairwise$pos_1 - null_pairwise$pos_2)
# Round r2 value to 3 decimal points
null_pairwise_df[[i]] <- null_pairwise %>% mutate(R2 = round(R2, 3))
null_pairwise_df[[i]] <- null_pairwise %>% mutate(Dprime = round(Dprime, 3))
# Used to build null distruibution of mean R2 for SNP-SNP pairs on diff
# chromosomes per sample 'pathway'
null_diff_chr[[i]] <- filter(null_pairwise_df[[i]], chr_1 != chr_2) %>%
dplyr::select(R2) %>% unlist
cat(" done.\n")
}
all_null <- do.call("rbind", null_pairwise_df)
cat(sprintf("Calculated %i randomly permuted SNP interactions.\n",
nrow(all_null)))
null_diff_num <- sapply(null_diff_chr, length)
#get mean r2 value per permutation
null_diff_mean <- sapply(null_diff_chr, mean)
rm(test)
#============================================================================#
## PLOT STATS
message("\n-------PLOTS-------\n")
cat("Linkage disequilbrium statistic plots...")
####### high confidence
#high confidence pairwise SNPs
all_top$pathway_group <- "highconf"
sub_top <- subset(all_top, R2 > 0.2)
#filter based on matching/non-matching chromosome pairs
all_top_same_chr <- filter(all_top, chr_1 == chr_2)
all_top_diff_chr <- filter(all_top, chr_1 != chr_2)
sub_top_same_chr <- filter(sub_top, chr_1 == chr_2)
#write out tables
write.table(all_top_same_chr,
file=sprintf("%s/all_top_same_chr.txt", outDir),
col=T, row=F, quote=F, sep="\t")
write.table(all_top_diff_chr,
file=sprintf("%s/all_top_diff_chr.txt", outDir),
col=T, row=F, quote=F, sep="\t")
####### random
#randomly selected pairwise SNPs
#mean R2 per sample random 'pathway' to build null distribution
#each row corresponds to each 'pathway' (nrow(null_diff_mean)==rep.num)
all_null$pathway_group <- "null"
sub_null <- subset(all_null, R2 > 0.2)
#filter based on matching/non-matching chromosome pairs
all_null_same_chr <- filter(all_null, chr_1 == chr_2)
all_null_diff_chr <- filter(all_null, chr_1 != chr_2)
sub_null_same_chr <- filter(sub_null, chr_1 == chr_2)
####### all
#combine high-conf and null dataframes containing pairwise R2 values
all_same_chr <- rbind(all_top_same_chr, all_null_same_chr)
all_diff_chr <- rbind(all_top_diff_chr, all_null_diff_chr)
sub_same_chr <- rbind(sub_top_same_chr, sub_null_same_chr)
#divide distance axes to make them human-readable
all_same_chr$dist <- all_same_chr$dist/1000000 ##divide by 1000000 (bp -> mb)
all_diff_chr$dist <- all_diff_chr$dist/1000000
sub_same_chr$dist <- sub_same_chr$dist/1000000
##############################################################################
title_same <- paste("Degree of corrrelation per intrachromosomal SNP-SNP",
"\npair within the high-confidence pathway group vs.",
"\nrandomly selected SNP groups")
title_diff <- paste("Degree of co-selection per interchromosomal SNP-SNP",
"\ninteraction within the high-confidence pathway",
"\ngroup vs. randomly selected SNP groups")
## for use with stat_summary(fun.data=box.style); allows white median line to
## appear after colouring and filling boxplots
box.style <- function(x){
return(c(y=median(x), ymin=median(x), ymax=median(x)))
}
## for use with stat.summary(fun.data=give.n); displays sample size (N)
## courtesy of Bangyou at Stack Overflow
give.n <- function(x){
return(c(y = median(x)*1.30, label = length(x)))
# experiment with the multiplier to find the perfect position
}
# boxplot of r2 per high confidence pathway
blah <- melt(top_diff_chr)
ggplot(blah, aes(x=factor(L1), y=value, color="#F8766D", fill="#F8766D")) +
geom_boxplot() +
stat_summary(geom="crossbar", width=0.65, fatten=0, color="white",
fun.data=box.style) +
stat_summary(geom="text", color="white", fun.data=give.n,
position=position_dodge(width=0.75)) +
scale_y_continuous(name=bquote("Pairwise LD value (" *r^2*")")) +
scale_x_discrete(name="# of interchromosomal SNP-SNP pairs per pathway") +
ggtitle(paste("Degree of co-selection per interchromosomal SNP-SNP",
"\npair within the high-confidence pathways")) +
theme(axis.text.x=element_text(vjust=0.4, hjust=1)) +
theme_set(theme_minimal()) +
theme(plot.title=element_text(hjust=0.5),
text=element_text(size=17),
legend.position="none",
panel.grid.major.x=element_blank()) +
geom_hline(yintercept=0.1, colour="grey", linetype="dashed", size=1)
ggsave("high_interchr_bar.png", width=11)
#################### PLOT 1: Null distribution vs real #######################
# Plotting mean R2 value for random SNP-SNP pairs on diff chromosome
# against real mean R2 value for high-conf SNPs
# interchromosomal SNP pairs used as proxy for co-selection/genetic ixn
null <- null_diff_mean # mean r2 values per permutation (n=500)
real <- mean(all_top_diff_chr$R2) # mean r2 for all high-conf pathways (n=1)
ggplot(as.data.frame(null), aes(x=as.data.frame(null))) +
geom_histogram(aes(y=..density..),
colour="white", fill="grey", bins=15) +
geom_density(alpha=0.3, colour="#00BFC4", fill="#00BFC4") +
scale_x_continuous(name=bquote("Pairwise LD value (mean " *r^2*")")) +
scale_y_continuous(name="Density") +
ggtitle(title_diff) +
#xintercept = mean R2 for high-conf SNP-SNP pairs on diff chrs (real R2)
geom_vline(aes(xintercept=real),
color="#F8766D", linetype="dashed", size=1) +
theme_set(theme_minimal()) +
theme(plot.title=element_text(hjust=0.5),
text=element_text(size=17),
legend.position="top",
legend.title=element_blank(),
panel.grid.major.x=element_blank())
ggsave("dist_null_real.png", width=8, height=7)
#calculate pvalue by quantifying all permuted mean r2 values greater than
#the real mean r2 values, divided by the total number of replicates
pval <- function(null, real) {
perm.p = ( length(which(null > mean(real)))+1 ) / (length(null)+1)
return(perm.p)
}
perm.p = pval(null, real)
cat(sprintf("P-value for permuted sample vs. real mean r2 = %g\n", perm.p))
########################## PLOT 2: R2 boxplots ###############################
# R2 boxplot without distance bins
dat <- sub_same_chr
ggplot(dat, aes(x=pathway_group, y=-log10(R2))) +
geom_boxplot(outlier.colour=NULL, aes(colour=pathway_group,
fill=pathway_group)) +
stat_summary(geom="crossbar", width=0.65, fatten=0, color="white",
fun.data=box.style) +
scale_y_continuous(name=bquote("Pairwise LD value (-log10("*r^2*"))")) +
#scale_y_continuous(name=bquote("Pairwise LD value ("*r^2*")")) +
ggtitle(title_same) +
theme_set(theme_minimal()) +
theme(plot.title=element_text(hjust=0.5),
text=element_text(size=17),
legend.position="top",
legend.title=element_blank(),
panel.grid.major.x=element_blank(),
axis.title.x=element_blank()) +
scale_x_discrete(labels=paste("N=", table(dat$pathway_group), sep=""))
ggsave("sub_same_log.png", width=7.5, height=7.5)
######################### PLOT 3: Density plots ##############################
# Get mean R2 for each pathway group and plot density
##NOTE:only do this for intrachromosomal SNPs (all_same_chr + sub_same_chr)
meanvals <- ddply(dat, "pathway_group", summarise, r2_mean=mean(R2))
ggplot(dat, aes(x=R2)) +
geom_density(aes(group=pathway_group, colour=pathway_group,
fill=pathway_group), alpha=0.3) +
scale_x_continuous(name=bquote("Pairwise LD value (" *r^2*")")) +
scale_y_continuous(name="Density") +
ggtitle(title_same) +
theme_set(theme_minimal()) +
theme(plot.title=element_text(hjust=0.5),
text=element_text(size=17),
legend.position="top",
legend.title=element_blank(),
panel.grid.major.x=element_blank()) +
geom_vline(data=meanvals, aes(xintercept=r2_mean,
colour=pathway_group), linetype="dashed", size=1)
ggsave("sub_density.png", width=8, height=7)
# Get significance values
dat <- all_diff_chr
wilcox <- wilcox.test(R2 ~ pathway_group, data=dat, alternative="greater")
ttest <- t.test(dat$R2 ~ pathway_group, var.equal=T)
print(wilcox)
print(ttest)
######################## PLOT 5: Linear regression ###########################
# Plot linear regression of R2 vs distance for both pathway groups
##NOTE:only regressing intrachromosomal SNPs (all_same_chr + sub_same_chr)
ggplot(dat, aes(x=dist, y=R2, colour=pathway_group)) +
geom_point() +
geom_smooth(method=lm, se=T) +
scale_y_continuous(name=bquote("Pairwise LD value (" *r^2*")")) +
scale_x_continuous(name="Chromosomal distance (Mb)") +
ggtitle(title_same) +
theme_set(theme_minimal()) +
theme(plot.title=element_text(hjust=0.5),
text=element_text(size=17),
legend.position="top",
legend.title=element_blank(),
panel.grid.major.x=element_blank()) +
coord_cartesian(xlim=c(0, 0.25), ylim=c(0,1)) #zoom in
ggsave("sub_regr_zoom.png", width=8, height=7)
# Determine significance of regression
fit <- lm(formula=R2 ~ dist + pathway_group, data=dat)
summary(fit)
#==============================================================================
# Plots no longer used
# R2 boxplot by distance (first cut distance into 4 equally sized groups)
# (q<-quantile(dat$dist, seq(0, 1, 0.25)))
# dat$dist_ranges <- cut(dat$dist, breaks=q, include.lowest=T)
# ggplot(dat, aes(x=dist_ranges, y=-log10(R2), colour=pathway_group,
# fill=pathway_group)) +
# geom_boxplot(outlier.colour=NULL) +
# stat_summary(geom="crossbar", width=0.65, fatten=0, color="white",
# fun.data=box.style) +
# scale_x_discrete("Chromosomal distance (Mb)") +
# scale_y_continuous(name=bquote("-log10("*R^2*")")) +
# ggtitle(title_same) +
# theme_set(theme_minimal()) +
# theme(plot.title=element_text(hjust=0.5), text=element_text(size=17),
# legend.position="top", legend.title=element_blank(),
# panel.grid.major.x=element_blank())
##contour plots - clearer visaul of bimodality
# ggplot(sub_same_chr, aes(dist, R2)) +
# geom_density_2d() +
# geom_point() +
# scale_y_continuous(name=expression(paste("R"^"2"))) +
# scale_x_continuous(name="Distance (bp)") +
# ggtitle(paste("Distribution of pairwise R2 measure as a function of",
# "distance\n within the high-confidence pathway group")) +
# geom_hline(aes(yintercept=0.2), colour="red") +
# theme(plot.title=element_text(hjust=0.5), text=element_text(size=17))
#Plot R2 for all high-conf pathways
##fill vectors with NA before melting list
# max_length <- max(unlist(lapply(top_ld_r2, length)))
# r2_filled <- lapply(top_ld_r2, function(x){ans <- rep(NA, length=max_length);
# ans[1:length(x)]<- x;
# return(ans)})
# r2_filled <- do.call(cbind, r2_filled)
# r2_melt <- melt(r2_filled)
##rename columns for better plot visualization
# colnames(r2_melt) <- c("SNP_pairs", "pathway", "R2")
# r2_melt <- na.omit(r2_melt)
# r2_melt$pathway <- paste("High-conf_path", r2_melt$pathway, sep=" ")
# ggplot(r2_melt, aes(x=SNP_pairs, y=R2)) +
# facet_wrap(~pathway, nrow=3, ncol=6) +
# scale_y_continuous(name=expression(paste("R"^"2"))) +
# scale_x_continuous(name="Number of SNP-SNP pairs in pathway") +
# geom_point(aes(colour=pathway)) +
# geom_hline(yintercept=0.5, colour="red") +
# theme_bw() +
# theme(text=element_text(size=15))
#ggsave()
}
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.