DEseq_gene_eRNA_30min.R

setwd("~/Documents/IFN_enhancer/R/")
library(GenomicRanges)
library(GenomicFeatures)
library(edgeR)
library(DESeq2)
library(cluster)
library(gplots)
source("~/Documents/IFN_enhancer/R/Script/ref_2_symbol_matrix.R")

load("Data/DEseq_input.RData")
# load("eRNA_expression_GM.RData")
# load("gene.rpkm.symbol.RData")

###############################################################
# 1. Perform DEseq
###############################################################
deseq_padj_cutoff <- 0.01
de_times_cutoff <- 1

deg_list <- NULL
deg_ct <- NULL
tp <- unlist(strsplit(colnames(countData), "_"))[c(0:11 * 4 + 1)]
groups <- factor(rep(tp, each = 2))
colData <- data.frame(groups = groups, row.names = colnames(countData))
dds <- DESeqDataSetFromMatrix(
  countData = countData,
  colData = colData,
  design = ~groups
)
dds <- dds[ rowSums(rpkmData > 1) > 1, ] # Pre-filtering
dds <- DESeq(dds, betaPrior = FALSE)
norm.counts <- counts(dds, normalized = T)

deg_matrix <- matrix(0, nrow = nrow(countData), ncol = 10, dimnames = list(rownames(countData), as.character(tp)[3:12]))
for (i in 3:12) {
  res <- results(dds, contrast = c("groups", as.character(tp[i]), "GM30min"))
  lab <- rownames(norm.counts)[which(res$padj < deseq_padj_cutoff)]
  deg_matrix[lab, i - 2] <- 1
}

de_all <- rownames(deg_matrix)[rowSums(deg_matrix[, 1:8]) > de_times_cutoff]
lab <- which(substring(de_all, 1, 12) == "MetaEnhancer")
de_genes <- de_all[-lab]
de_eRNAs <- de_all[lab]

save(norm.counts, countData, deg_matrix, file = "Data/deg_matrix_GM_30min.RData")

###############################################################
# 2. Some stats for genes and for eRNAs
###############################################################

load(file = "Data/deg_matrix_GM_30min.RData")

deg_matrix <- deg_matrix[rownames(deg_matrix) %in% rownames(norm.counts), ]
deg_matrix_direction <- deg_matrix
ref_col <- c(3:4)
initial <- log2(rowMeans(norm.counts[, ref_col] + 1))
for (i in 1:ncol(deg_matrix)) {
  lab <- which(deg_matrix[, i] == 1)
  a <- log2(rowMeans(norm.counts[lab, i * 2 + ref_col]) + 1)
  b <- initial[lab]
  deg_matrix_direction[lab[a - b < 0], i] <- -1
}
count_deg <- function(x) {
  return(c(
    sum(x == 1), -sum(x == -1)
  ))
}
tp <- apply(deg_matrix_direction, 2, count_deg)

barplot(tp[1, ],
  ylim = c(-1000, 1000), col = "red", ylab = "Number of Differentially Expressed Genes",
  names.arg = substr(colnames(tp), 3, 1000L)
)
barplot(tp[2, ], add = T, col = "blue", names.arg = F)


###############################################################
# 3. Summarize gene expression profiles near differentially expressed eRNAs
###############################################################

x <- eRNA.rpkm[, 2:10]
lab <- de_eRNAs[de_eRNAs %in% rownames(x)]
x <- x[lab, ]
x <- (x - rowMeans(x)) / apply(x, 1, sd)

# Sort rows by kmeans clustering
zscore <- x
set.seed(0)
centers <- 3
kmeans <- kmeans(zscore, centers, iter.max = 10^5)

k_cluster <- kmeans$cluster
cluster_order <- c(2, 1, 3)
ind <- cluster_order[k_cluster] * 10^4 + rowMeans(zscore)
od <- order(ind, decreasing = T)
zscore <- zscore[od, ]
k_cluster <- k_cluster[od]

# eRNA heatmap
my_palette <- colorRampPalette(c("blue", "black", "yellow"), bias = 1)(n = 500)
heatmap.2(zscore[nrow(zscore):1, ],
  col = my_palette, Colv = F, Rowv = F, dendrogram = "none",
  labRow = NA, keysize = 1.5, density.info = "density",
  key.ylab = "", key.xlab = "", key.title = "", trace = "none"
)

par(mfcol = c(2, 2), mar = c(4, 4, 4, 5))
for (i in cluster_order) {
  boxplot(zscore[k_cluster == i, ], ylab = "Zscore")
}

###############################################################
# 4. Finding correlated eRNA-gene pairs
###############################################################

tss <- promoters(gene, upstream = 1, downstream = 1)
flanking <- 200000
points_to_consider <- c(1:10)
e_mat <- eRNA.rpkm[, points_to_consider]
g_mat <- gene.rpkm.symbol[, points_to_consider]

find_target <- function(x) {
  tp <- eRNA[mcols(eRNA)$id == x]
  start(tp) <- start(tp) - flanking
  end(tp) <- end(tp) + flanking
  hits <- findOverlaps(tp, tss, ignore.strand = T)
  lab <- as.character(unique(mcols(tss[subjectHits(hits)])$id))
  lab <- ref_2_symbol(lab)
  lab <- lab[lab %in% de_genes]
  return(lab)
}

find_region <- function(x, y) {
  a <- x
  b <- y
  chr <- as.character(seqnames(a))
  position <- c(start(a), end(a), start(b), end(b))
  dist <- abs(mean(c(start(a), end(a))) - start(promoters(b))[1])
  dist <- ceiling(dist / 1000)
  left <- min(position) - 2000
  right <- max(position) + 2000
  return(paste(chr, " ", left, " ", right, " (", dist, "kb)", sep = ""))
}

pdf(file = "Correlated_eRNA_gene_pairs_GM.pdf", height = 9, width = 9)
par(mfrow = c(3, 3))
my_palette <- colorRampPalette(c("blue", "red"), bias = 1)(n = 9)
for (i in cluster_order) {
  eRNA_list <- rownames(zscore)[k_cluster == i]
  for (j in 1:length(eRNA_list)) {
    lab <- find_target(eRNA_list[j])
    if (length(lab) > 0) {
      for (k in 1:length(lab)) {
        pvalue <- cor.test(e_mat[eRNA_list[j], ], g_mat[lab[k], ])$p.value
        region <- find_region(eRNA[mcols(eRNA)$id == eRNA_list[j]], gene[mcols(gene)$id %in% sym_2_ref(lab[k])])
        if (pvalue < 0.01) {
          plot(e_mat[eRNA_list[j], ], g_mat[lab[k], ],
            col = my_palette, pch = 16, cex = 1.5,
            main = region, xlab = eRNA_list[j], ylab = lab[k]
          )
          legend("topleft", paste("P <", signif(pvalue, 1)), bty = "n")
        }
      }
    }
  }
}
dev.off()

###############################################################
# 5. Others
###############################################################
load("deg_matrix_GM_compare_with_30min.RData")
de_genes <- rownames(deg_matrix)[rowSums(deg_matrix[, 1:8]) > 0]
tss <- promoters(gene, upstream = 1, downstream = 1)
flanking <- 500000
for (i in cluster_order) {
  lab <- rownames(zscore)[k_cluster == i]
  tp <- eRNA[mcols(eRNA)$id %in% lab]
  start(tp) <- start(tp) - flanking
  end(tp) <- end(tp) + flanking
  hits <- findOverlaps(tp, tss)
  lab <- as.character(unique(mcols(tss[subjectHits(hits)])$id))
  lab <- ref_2_symbol(lab)
  lab <- lab[lab %in% de_genes]
  if (i == 2) {
    write.table(lab, quote = F, row.names = F, col.names = F, file = "test.txt")
  }
  tp <- gene.rpkm.symbol[lab, ]
  tp <- (tp - rowMeans(tp)) / apply(tp, 1, sd)
  print(length(lab))
  # boxplot(tp, ylab="Zscore")
  heatmap.2(tp,
    col = my_palette, Colv = F, Rowv = T, dendrogram = "none",
    keysize = 1.5, density.info = "density",
    key.ylab = "", key.xlab = "", key.title = "", trace = "none"
  )
}

par(mfrow = c(1, 2))
my_palette <- colorRampPalette(c("blue", "red", "orange"), bias = 1)(n = 9)
plot(eRNA.rpkm["MetaEnhancer_65_-", 2:10], gene.rpkm.symbol["IRF2", ],
  col = my_palette, pch = 16, cex = 1.3,
  xlab = "MetaEnhancer_65_-", ylab = "IRF2"
)
plot(eRNA.rpkm["MetaEnhancer_250_-", 2:10], gene.rpkm.symbol["IRF1", ],
  col = my_palette, pch = 16, cex = 1.3,
  xlab = "MetaEnhancer_250_-", ylab = "IRF1"
)

plot(eRNA.rpkm["MetaEnhancer_116_+", 2:10], gene.rpkm.symbol["TNFSF10", ],
  col = my_palette, pch = 16, cex = 1.3,
  xlab = "MetaEnhancer_116_+", ylab = "TNFSF10"
)