METAL450k.Rmd

---
title: "METAl meta-Analysis"
author: "The methylation in ART meta-analysis group"
date: "10/1/2020"
output:   
  html_document:
    toc: true
    fig_width: 7
    fig_height: 7
theme: cosmo
---

# Introduction
Here we run the possible meta analysis comparisons

Match anno to probes
DMRcate peak calling
pvalue histogram
volcano plot
manhattan plot with split axis (split by z score) plot hypo hyper separately
Effectsize Deltabeta
Forrest Plot

bumpHunter, DMRfinder, probelasso, mCSEA, DMRforPairs 


```{r, install packages}

source("meth_functions.R")
ann450k <- getAnnotation(IlluminaHumanMethylation450kanno.ilmn12.hg19)
myann <- data.frame(ann450k[,c("chr", "pos", "UCSC_RefGene_Name","Regulatory_Feature_Group","Islands_Name","Relation_to_Island")])
library(bumphunter)
library(zoo)

```

MARKER = Row.names
DEFAULT = N (Neff=4/(1/Ncases+1/Nctrls))
ALLELE = unmeth meth
FREQ = AveExpr
EFFECT = CE
STDERR = SE
PVAL = P.Value


# DMR Caller

```{r, DMR caller}

# K=number of probes in rolling average
# RAT=rolling average threshold
# data=limma result. do not filter any probes based on p-values
dmrcaller <- function(data,K,RAT){
  res <- data
  res <- makeGRangesFromDataFrame(res, keep.extra.columns = TRUE, ignore.strand = TRUE,
    seqnames.field = "chr", start.field = "pos", end.field = "pos")
  res <- sortSeqlevels(res)
  res <- sort(res)
  resdf <- as.data.frame(res)
  temp <- as.vector(apply(resdf, 1, function(x){
    if (as.numeric(x[11]) > 0.05){ y <- 0 }
    else if (as.numeric(x[10])> 0){ y<- 1 }
    else if (as.numeric(x[10])< 0){ y<- -1 }
    y
  }))
  rolling <- rollmean(temp,k=K, fill = NA)
  resdf$temp <- temp
  resdf$rolling <- rolling
  # hyer
  hyper <- as.data.frame(t(sapply(which(rolling >= RAT), function(x){
    w=x-1
    y=x+1
    dmrstart <- start(res)[w]
    dmrend <- start(res)[y]
    chr <- seqnames(res)[y]
    return(c(w, x, y, dmrstart, dmrend))
  })))
  hyper_chr <- sapply(which(rolling >= RAT), function(x){
    chr <- seqnames(res)[x]
    as.character(chr)
  })
  hyper$chr <- hyper_chr
  colnames(hyper)<- c("w", "x", "y", "start", "end", "chr")
  hyperGR <- makeGRangesFromDataFrame(hyper, keep.extra.columns = TRUE,
    ignore.strand = TRUE, seqnames.field = "chr", start.field = "start", end.field = "end")
  hyperGR <- GenomicRanges::reduce(hyperGR)
  # extract the probes that overlap each DMR
  hyperOL <- apply(as.data.frame(hyperGR), 1, function(x){
    G <- GRanges(seqnames = x[1], ranges = IRanges(x[2]:x[3]))
    subsetByOverlaps(res, G)
  })
  # from metaRNA-seq CRAN pkg
  fishercomb <- function(indpval) {
    logpval=do.call(cbind,lapply(indpval, log))
    statc=apply(logpval,1, FUN=function(x) -2*sum(x,na.rm=TRUE))
    na.index <- which(apply(logpval, 1, function(x) sum(is.na(x))) == ncol(logpval))
    statc[na.index] <- NA
    notNA=apply(logpval,1,FUN=function(x) sum(!(is.na(x))))
    rpvalc = 1 - pchisq(statc, df=(2*notNA))
    return(rpvalc)
  }
  # obtain summary info for each hyper DMR
  hyperdmr <- lapply(hyperOL,function(x) {
    df <- as.data.frame(x)
    n <- nrow(df)
    z <- mean(df$Zscore)
    fp <- fishercomb(df$P.value)
    chr <- df[1,1]
    start <- df[1,2]
    end <- df[nrow(df),2]
    data.frame("chr"=chr,"start"=start,"end"=end,"no. probes"=n,"meanZ"=z,"FisherP"=fp)
  })
  hyperdmr <- do.call(rbind, hyperdmr)
  # hypo
  hypo <- as.data.frame(t(sapply(which(rolling <= -RAT), function(x){
    w=x-1
    y=x+1
    dmrstart <- start(res)[w]
    dmrend <- start(res)[y]
    chr <- seqnames(res)[y]
    return(c(w, x, y, dmrstart, dmrend))
  })))
  hypo_chr <- sapply(which(rolling <= -RAT), function(x){
    chr <- seqnames(res)[x]
    as.character(chr)
  })
  hypo$chr <- hypo_chr
  colnames(hypo)<- c("w", "x", "y", "start", "end", "chr")
  hypoGR <- makeGRangesFromDataFrame(hypo, keep.extra.columns = TRUE,
    ignore.strand = TRUE, seqnames.field = "chr", start.field = "start", end.field = "end")
  hypoGR <- GenomicRanges::reduce(hypoGR)
  # extract the probes that overlap each DMR
  hypoOL <- apply(as.data.frame(hypoGR), 1, function(x){
    G <- GRanges(seqnames = x[1], ranges = IRanges(x[2]:x[3]))
    subsetByOverlaps(res, G)
  })
  # obtain summary info for each hypo DMR
  hypodmr <- lapply(hypoOL,function(x) {
    df <- as.data.frame(x)
    n <- nrow(df)
    z <- mean(df$Zscore)
    fp <- fishercomb(df$P.value)
    chr <- df[1,1]
    start <- df[1,2]
    end <- df[nrow(df),2]
    data.frame("chr"=chr,"start"=start,"end"=end,"no. probes"=n,"meanZ"=z,"FisherP"=fp)
  })
  hypodmr <- do.call(rbind, hypodmr)

  list("HyperDMRs"=hyperdmr,"HypoDMRs"=hypodmr,"hyper overlap"=hyperOL,"hypo overlap"=hypoOL)
}

```



# Natural vs ART

Here we compare all data sets with by Natural vs ART, as all studies use different methods this is a 
general comparison.
Estill
Choufani
Litzky
Hajj

```{bash, Natural vs ART} 

metal NATvsART.txt

```

```{r, Natural vs ART, output}

nat_vs_art <- read.table("METAANALYSIS1.TBL", header = TRUE)
nat_vs_art <- nat_vs_art[order(nat_vs_art$P.value),]
nat_vs_art$FDR <- p.adjust(nat_vs_art$P.value)

nat_vs_art_a <- merge(nat_vs_art, myann, by.x = "MarkerName", by.y = 0)
nat_vs_art_a <- nat_vs_art_a[order(nat_vs_art_a$P.value),]
head(nat_vs_art_a, 20)

hist(-log10(nat_vs_art_a$FDR))
hist(-log10(nat_vs_art_a$P.value))
hist(nat_vs_art_a$FDR)
hist(nat_vs_art_a$P.value, main = "Histogram of P.values")

volcano_metal(nat_vs_art_a, "Nat vs ART")

nat_vs_art_a$chrnum <- as.integer(gsub("chr","", nat_vs_art_a$chr))

manhattan(nat_vs_art_a, chr="chrnum", bp="pos", p="P.value")
manhattan(nat_vs_art_a, chr="chrnum", bp="pos", p="FDR")

hyper <- subset(nat_vs_art_a, Zscore > 0)
hypo <- subset(nat_vs_art_a, Zscore < 0)

manhattan(hyper, chr="chrnum", bp="pos", p="P.value", )
manhattan(hypo, chr="chrnum", bp="pos", p="P.value")

```

```{r,dmrcalling}

# testing on chr22
#data1 <- nat_vs_art_a[which(nat_vs_art_a$chr=="chr22"),]
#system.time(dmrres <- dmrcaller(data=data1,K=5,RAT=0.7))

system.time(dmrres <- dmrcaller(data=nat_vs_art_a,K=5,RAT=0.7))

names(dmrres)
hyper <- dmrres$HyperDMRs
nrow(hyper)
head( hyper[order(hyper$FisherP),] ,20 )
hypo <- dmrres$HypoDMRs
nrow(hypo)
head( hypo[order(hypo$FisherP),] ,20 )


```

# Natural vs ICSI 
Estill
Choufani
Hajj

```{bash, Natural vs ICSI}

metal NATvsICSI.txt

```

```{r, Natural vs ICSI, output}
nat_vs_icsi <- read.table("METAANALYSIS1.TBL", header = TRUE)
nat_vs_icsi <- nat_vs_icsi[order(nat_vs_icsi$P.value),]
head(nat_vs_icsi, 20)
```


# Natural vs Subfertility
Two of the studies also listed subfertility as a parameter, here we compare the two studies.
Choufani
Litzky
```{bash, Natural vs SUB}

metal NATvsSUB.txt

```

```{r, Natural vs Sub, output}
nat_vs_sub <- read.table("METAANALYSIS1.TBL", header = TRUE)
nat_vs_sub <- nat_vs_sub[order(nat_vs_sub$P.value),]
head(nat_vs_sub, 20)
```
Note that the `echo = FALSE` parameter was added to the code chunk to prevent printing of the R code that generated the plot.


# Natural vs ART in Placenta
Two of the studies have samples taken from placenta both of which when analysed individually have few-no significant probes, here we compare the two studies.
Choufani
Litzky
```{bash, Natural vs ART in placenta}

metal NATvsARTplacenta.txt

```

```{r, Natural vs ART placenta, output} 
nat_vs_artp <- read.table("METAANALYSIS1.TBL", header = TRUE)
nat_vs_artp <- nat_vs_artp[order(nat_vs_artp$P.value),]
head(nat_vs_artp, 20)
```
Note that the `echo = FALSE` parameter was added to the code chunk to prevent printing of the R code that generated the plot.


## Session Information

```{r,session}

sessionInfo()

```