#GSE141941
#suppose all bw files in current directory
require(rtracklayer)
require(rTensor)
files <- list.files("./")
files <- files[intersect(grep("GSM",files),grep(".bw",files))]
j=1
dir <- gsub(".bw","",files[j])
system(paste("mkdir",dir))
which <- GRanges("chr1", IRanges(1, 30000))
bw <- import(files[j],which=which)
id <- c(1:length(bw@seqinfo@seqnames))[-grep("_",bw@seqinfo@seqnames)]
for (k in c(1:length(id)))
{
    seqname <- bw@seqinfo@seqnames[id[k]]
    if (bw@seqinfo@seqlengths[id[k]]>=25000)
    {
        breaks<-seq(1,bw@seqinfo@seqlengths[id[k]],by=25000)
    } else {
        breaks<- c(1,bw@seqinfo@seqlengths[id[k]])
    }
    Chip <- NULL
    for (i in (c(1:(length(breaks)-1))))
    {
        cat(i," ")
        which <- GRanges(seqname, IRanges(breaks[i], breaks[i+1]-1))
        bw <- import(files[j],which=which)
        Chip <- c(Chip,sum(bw$score))
    }
    save(file=paste(dir,"/Chip_",seqname,sep=""),Chip)
}

dirs <- list.files(pattern="GSM")
dirs <- dirs[-grep("bw",dirs)]
chr <- paste("chr",c(1:22,"M","X","Y"),sep="")
for (l in c(1:length(chr)))
{
    cat(l," ")
    x<-NULL
    for (i in c(1:length(dirs)))
    {
        #cat (i, " ")
        load(paste(dirs[i],"/Chip_",chr[l],sep=""))
        x <- cbind(x,Chip)
    }
    
    
    set.seed(0)
    Z <- array(NA,c(dim(x)[1],4,2))
    Z[,1,1] <- data.matrix(x[,1])
    Z[,1,2] <- data.matrix(x[,2])    
    Z[,2,1] <- data.matrix(x[,3])
    Z[,2,2] <- data.matrix(x[,4])   
    Z[,3,1] <- data.matrix(x[,5])
    Z[,3,2] <- data.matrix(x[,6])   
    Z[,4,1] <- data.matrix(x[,7])
    Z[,4,2] <- data.matrix(x[,8])   
    save(file=paste("Z_",chr[l],sep=""),Z)
}

require(abind)
dnum <-NULL
Z_all <- NULL
for(l in c(1:length(chr)))
{
    load(paste("./Z/Z_",chr[l],sep=""))
    dnum<-c(dnum,dim(Z)[1])
    Z_all <- abind(Z_all,Z,along=1)
}
#save(file="Z_all",Z_all)
#save(file="dnum",dnum)

HOSVD <- hosvd(as.tensor(Z_all),c(10,4,2))
save(file="HOSVD",HOSVD)
pdf(file="HOSVD.pdf",width=10,height=5)
par(mfrow=c(1,2))
matplot(c(0,6,12,24),HOSVD$U[[2]],type="o",pch=1:4,xlab="hours",ylab="PC loading",cex=2,lwd=2,cex.lab=1.5,cex.axis=2)
abline(0,0,col=6,lty=2,lwd=2)
plot(pca$rotation[,2],HOSVD$U[[2]][,2],xlab="gene expression",ylab="m6A",
     cex=2,xlim=c(-0.7,0.9),ylim=c(-0.7,0.8),cex.axis=2,cex.lab=1.5,lwd=2)
abline(0,1,col=2,lty=2,lwd=2)
par(mfrow=c(1,1))
dev.off()
cor(c(0,6,12,24),HOSVD$U[[2]])
HOSVD$Z@data[,2,2]
P <- pchisq(scale(HOSVD$U[[1]][,2])^2,1,lower.tail=F)
table(p.adjust(P,"BH")<0.01)
require(biomaRt)
grch37 = useMart(biomart="ENSEMBL_MART_ENSEMBL", host="grch37.ensembl.org", path="/biomart/martservice", dataset="hsapiens_gene_ensembl")

gene_all <-NULL
for (l in c(1:22)){
    cat("\n l= ",l,"\n")
    if (l==1)
    {
        id <- c(1:dnum[l])[(p.adjust(P,"BH")<0.01)[1:dnum[l]]]    
    }else{
        id <- c(1:dnum[l])[(p.adjust(P,"BH")<0.01)[sum(dnum[1:(l-1)])+c(1:dnum[l])]]    
    }
    if  (length(id)>0)
    {
        reg<-id*25000
        reg<-data.frame(reg-24999,reg)
        gene <- NULL
        for (i in c(1:dim(reg)[1]))
        {
            cat(i," ")
            gene_lst0 <- getBM(attributes=c('hgnc_symbol','entrezgene_id','chromosome_name','start_position','end_position'),filters = c('chromosome_name','start','end'), values=list(l,reg[i,1],reg[i,2]), mart = grch37)
            gene <- rbind(gene,gene_lst0)
        }
        gene <- gene[!is.na(gene[,1]),]
        gene <- gene[!is.na(gene[,1]) & gene[,1]!="",]
        gene_all <- rbind(gene_all,gene)
    }
}

#save(file="gene_all",gene_all)
write.table(file="gene_all.csv",gene_all,row.names=F,col.names=F,quote=F,sep="\t")
x <- read.xlsx("GSE141941_normoxiaVShypoxia6h.12h.24h_RNA-seq.PROCESSED.DATA.xlsx",sheetIndex=1)
#save(file="x",x)
pca <- prcomp(scale(x[,-1]))
pdf(file="PCA.pdf",width=5,height=5)
matplot(c(0,6,12,24),pca$rotation,type="o",pch=1:4,xlab="hours",ylab="PC loading",cex=2,lwd=2,cex.lab=1.5,cex.axis=2)
abline(0,0,col=6,lty=2,lwd=2)
dev.off()

cor(c(0,6,12,24),pca$rotation)
P <- pchisq(scale(pca$x[,2])^2,1,lower.tail=F)
table(p.adjust(P,"BH")<0.01)

data.frame(unlist(lapply(strsplit(x[p.adjust(P,"BH")<0.01,1],".",fixed=T),"[",1)))

#PCA method 
data.frame(unlist(lapply(strsplit(as.character(x[order(-abs(pca$x[,1]))[1:52],1]),".",fixed=T),"[",1)))

gene<- read.ods("gene_RNA-seq.ods",sheet=2)
load("gene_all")
table(unique(gene_all[,1]) %in% gene[,2])
fisher.test(matrix(c(20000-189-45+7,189,45,7),2))
ZZ <- tensor(as.array(Z_all),as.array(Z_all),1,1)
xx <- t(data.matrix(x[,-1])) %*% data.matrix(x[,-1])
Zx <- tensor(ZZ,xx,1,1)
HOSVD <- hosvd(as.tensor(Zx))

pdf(file="TKD.pdf",width=15,height=5)
par(mfrow=c(1,3))
matplot(c(0,6,12,24),HOSVD$U[[4]],type="o",pch=1:4,xlab="hours",
        ylab="singlar value vectors",cex=2,lwd=2,cex.lab=1.5,cex.axis=2,main="gene expression")
abline(0,0,col=6,lty=2,lwd=2)
matplot(c(0,6,12,24),HOSVD$U[[2]],type="o",pch=1:4,xlab="hours",
        ylab="singlar value vectors",cex=2,lwd=2,cex.lab=1.5,cex.axis=2,main="m6A")
abline(0,0,col=6,lty=2,lwd=2)
plot(HOSVD$U[[4]][,2],HOSVD$U[[2]][,2],xlab="gene expression",ylab="m6A",
     cex=2,xlim=c(-0.7,0.9),ylim=c(-0.7,0.8),cex.axis=2,cex.lab=1.5,lwd=2)
abline(0,1,col=2,lty=2,lwd=2)
par(mfrow=c(1,1))
dev.off()

cor(c(0,6,12,24),HOSVD$U[[4]])
cor(c(0,6,12,24),HOSVD$U[[2]])
v<-tensor(tensor(Z_all,HOSVD$U[[1]][,2],3,1),HOSVD$U[[2]][,2],2,1)

P <- pchisq(scale(v)^2,1,lower.tail=F)
table(p.adjust(P,"BH")<0.01)
#save(file="gene_all_Z",gene_all)
write.table(file="gene_all_Z.csv",gene_all,row.names=F,col.names=F,quote=F,sep="\t")

load("gene_all_Z")
gene_all_Z <- gene_all
load("gene_all")

table(unique(gene_all[,1]) %in% unique(gene_all_Z[,1]))

u <- data.matrix(x[,-1]) %*% HOSVD$U[[4]][,2]
P <- pchisq(scale(u)^2,1,lower.tail=F)
table(p.adjust(P,"BH")<0.01)

data.frame(unlist(lapply(strsplit(x[p.adjust(P,"BH")<0.01,1],".",fixed=T),"[",1))) #gene_RNA-seq_Z

gene<- read.ods("gene_RNA-seq.ods",sheet=2)
gene1<- read.ods("gene_RNA-seq_Z.ods",sheet=2)

table(gene[,2] %in% gene1[,2])

load("gene_all_Z")
table(unique(gene_all[,1]) %in% gene1[,2])

fisher.test(matrix(c(20000-188-41+12,188,41,12),2))

LM <- lm(t(scale(x[,-1]))~c(0,6,12,24))
SLM <- summary(LM)
fs <- lapply(SLM,"[",10)
fs <- t(data.frame(lapply(SLM,"[",10)))
P <- pf(fs[,1],fs[,2],fs[,3],lower.tail=F)
table(p.adjust(P,"BH")<0.01)

table(p.adjust(P,"BH")<0.05)

SAM <- sam(scale(x[,-1]),0:3)

require(limma) 
design <- model.matrix(~0+c(1,2,3,4))
colnames(design) <- c("N")
fit <- lmFit(log(x[,-1]), design)
fit <- eBayes(fit)
print(table(topTable(fit,num=dim(x[,-1])[1])[,5]<0.01))

print(table(topTable(fit,num=dim(x[,-1])[1])[,5]<0.05))

dim(Z_all) <- c(dim(Z_all)[1],8)
#LM <- lm(t(Z_all)~as.vector(outer(c(0,6,12,24),1:2)))
#LM <- lm(t(Z_all)~as.vector(outer(c(0,6,12,24),2:1)))
SLM <- summary(LM)
fs <- lapply(SLM,"[",10)
fs <- t(data.frame(lapply(SLM,"[",10)))
P <- pf(fs[,1],fs[,2],fs[,3],lower.tail=F)
table(p.adjust(P,"BH")<0.01)

require(limma) 
design <- model.matrix(~0+as.vector(outer(c(1:4),c(0,4),"+"))) 
colnames(design) <- c("N")
fit <- lmFit(log(Z_all), design)
fit <- eBayes(fit)
print(table(topTable(fit,num=dim(Z_all)[1])[,5]<0.01))

print(table(topTable(fit,num=dim(Z_all)[1])[,5]<0.05))


set.seed(0)
rf <- randomForest(t(scale(x[,-1])),c(1:4))
table(abs(rf$importance)>0)

data.frame(unlist(lapply(strsplit(x[abs(rf$importance)>0,1],".",fixed=T),"[",1))) #gene_RNA-seq_rf

set.seed(0)
rf <- randomForest(t(Z_all),as.vector(outer(c(1:4),1:2)))
table(abs(rf$importance)>0)

require(biomaRt)
grch37 = useMart(biomart="ENSEMBL_MART_ENSEMBL", host="grch37.ensembl.org", path="/biomart/martservice", dataset="hsapiens_gene_ensembl")

gene_all <-NULL
for (l in c(1:22)){
    cat("\n l= ",l,"\n")
    if (l==1)
    {
        id <- c(1:dnum[l])[(abs(rf$importance)>0)[1:dnum[l]]]    
    }else{
        id <- c(1:dnum[l])[(abs(rf$importance)>0)[sum(dnum[1:(l-1)])+c(1:dnum[l])]]    
    }
    if  (length(id)>0)
    {
        reg<-id*25000
        reg<-data.frame(reg-24999,reg)
        gene <- NULL
        for (i in c(1:dim(reg)[1]))
        {
            cat(i," ")
            gene_lst0 <- getBM(attributes=c('hgnc_symbol','entrezgene_id','chromosome_name','start_position','end_position'),filters = c('chromosome_name','start','end'), values=list(l,reg[i,1],reg[i,2]), mart = grch37)
            gene <- rbind(gene,gene_lst0)
        }
        gene <- gene[!is.na(gene[,1]),]
        gene <- gene[!is.na(gene[,1]) & gene[,1]!="",]
        gene_all <- rbind(gene_all,gene)
    }
}

#save(file="gene_all_rf",gene_all)
write.table(file="gene_all_rf.csv",gene_all,row.names=F,col.names=F,quote=F,sep="\t")

LARS <- lars(t(scale(x[,-1])),c(1:4))
There are more than 500 variables and n<m;
You may wish to restart and set use.Gram=FALSE
table(LARS$beta[4,]!=0)

LARS <- lars(t(Z_all),as.vector(outer(c(1:4),1:2)))
There are more than 500 variables and n<m;
You may wish to restart and set use.Gram=FALSE
table(LARS$beta[8,]!=0)

#GSE120860
files <- list.files("./",pattern="bed.gz")
seqnames_all <- NULL
for (i in c(1:length(files)))
{
    cat(i," ")
    bed <- read.csv(files[i],sep="\t",header=F)
    seqnames_all <- c(seqnames_all,bed[,1])
}
seqnames_all <- unique(seqnames_all)
x_all <-NULL
for (i in c(1:length(files)))
{
    cat(i," ")
    bed <- read.csv(files[i],sep="\t",header=F)
    if (i <= 8)
    {
        SP<-split(bed[,4],bed[,1])
    } else {
        SP<-split(bed[,7],bed[,1])
    }
    SP <-unlist(lapply(SP,sum))
    vec <- rep(0,length(seqnames_all))
    vec[match(names(SP),seqnames_all)]<- SP
    x_all <- cbind(x_all,vec)
}
x_all <- data.frame(seqnames_all,x_all)
colnames(x_all)[-1] <- gsub(".bed.gz","",gsub("GSE120860_","",files))
#save(file="x_all",x_all)
Z <- array(NA,c(dim(x_all)[1],4,2,2))
Z[,1,1,1] <- x_all[,2] #0204.13P
Z[,1,1,2] <- x_all[,3] #0204.13T
Z[,2,1,1] <- x_all[,4] #0204.2P
Z[,2,1,2] <- x_all[,5] #0204.2T
Z[,3,1,1] <- x_all[,6] #0204.4P
Z[,3,1,2] <- x_all[,7] #0204.4T
Z[,4,1,1] <- x_all[,8] #0204.5P
Z[,4,1,2] <- x_all[,9] #0204.5T
Z[,1,2,1] <- x_all[,12] #patient.1P
Z[,1,2,2] <- x_all[,13] #patient.1T
Z[,2,2,1] <- x_all[,14] #patient.2P
Z[,2,2,2] <- x_all[,15] #patient.2T
Z[,3,2,1] <- x_all[,16] #patient.3P
Z[,3,2,2] <- x_all[,17] #patient.3T
Z[,4,2,1] <- x_all[,18] #patient.4P
Z[,4,2,2] <- x_all[,19] #patient.4T
Z <- apply(Z,2:4,scale)
P <- pchisq(scale(HOSVD$U[[1]][,4])^2,1,lower.tail=F)
table(p.adjust(P,"BH")<0.01)

data.frame(x_all[p.adjust(P,"BH")<0.01,1])