library(bioxm)

args <- commandArgs(TRUE)

outfile <- args[1]
sessionId <- args[2]
query <- args[3]
view <- args[4]

con <- bioxmConnect('../../bioxm2', sessionId=sessionId)
df <- getData(con, query, view, with.types=TRUE)
m <- as.matrix(df)

group.variable.1 <- unlist(strsplit(getQueryVariable(con, query, number=1), ";"))
group.variable.2 <- unlist(strsplit(getQueryVariable(con, query, number=2), ";"))

group.variable.id.1 <- sapply(strsplit(group.variable.1, ":"), function(x){x[3]})
group.variable.id.2 <- sapply(strsplit(group.variable.2, ":"), function(x){x[3]}) 

num.samples <- length(group.variable.1) + length(group.variable.2)

variable.names <- colnames(m);
new.variable.names <- c();
for (name in variable.names) {
  if (name %in% group.variable.id.1) {
    new.variable.names <- append(new.variable.names, paste(name, "(Group1)"));
  } else if (name %in% group.variable.id.2) {
    new.variable.names <- append(new.variable.names, paste(name, "(Group2)")); 
  } else {
    new.variable.names <- append(new.variable.names, name);
  }
}

colnames(m) <- new.variable.names;

#write.table(m, file=outfile, sep="\t", quote=FALSE)

loading.dir <- gsub('.out', '', outfile)
dir.create(loading.dir)


loading<-function(pca.result, pngfile) {
 
    par(mar=c(0,0,0,0))
    GDD(file = pngfile, type = "png", width = 150, height = 180, ps = 12, bg = "white")
    plot(pca.result$rotation[,2], type="h", xlab='', ylab='', font.lab=2, cex.lab=0.6, cex.axis = 0.6, cex=0.8)

    # define the threshold for the negative loadings:k.n
    t.n<-pca.result$rotation[which(pca.result$rotation[,2]<0),2]
    k<-quantile(abs(t.n),probs = seq(0, 1, 0.01))[95]
    k.n<-as.numeric(k[1])

    #the threshold for the positive loadings:k.p
    t.p<-pca.result$rotation[which(pca.result$rotation[,2]>0),2]
    k<-quantile(t.p,probs = seq(0, 1, 0.01))[95]
    k.p<-as.numeric(k[1])

    #plot the lines of the threshold
    abline(h=0)
    abline(h=k.p,col="red")
    abline(h=-k.n,col="green")
    dev.off()
}


pca.analysis <- function(input, output.file.name.PC1, output.file.name.PC2, number.of.samples){

	#input <- read.delim(input,row.names=NULL,header=T)
	#kegg.sub <- as.matrix(input)
        kegg.sub <- input	

	#adapt the output of BioXM for PCA
	#there are several pathways in one row,split it:
  
	stock.list <- c()
	#print("start formatting")
	for(i in seq(kegg.sub[,3])){
		
		kegg.tmp <- unlist(strsplit(as.character(kegg.sub[i,3]),";"))
		
		for(j in seq(kegg.tmp)){
			stock.list <- rbind(stock.list,c(kegg.tmp[j],kegg.sub[i,-3]))
		}
	}
	#print("the input is formatted")

	unique.kegg.pathway.list <- unique(stock.list[,1])

	#stocker for kegg pathways
	#stocker for the number of probes in each kegg pathway
	#stocker for the output files
	number.of.genes.stocker <- c()
	kegg.name.stocker <- c()
	final.pca.output.stocker.PC1 <- c()
	final.pca.output.stocker.PC2 <- c()
        loadings.images <- c()

	#pca analysis
	#for each kegg pathway calulates the PC1 and PC2

	for( i in seq(unique.kegg.pathway.list)){
		in.pathway <- which( stock.list[,1] %in% unique.kegg.pathway.list[i], arr.ind=T)

		# if there are more than 9 genes in a patwhay, we calculate the PC1 and PC2
		if (length(in.pathway) > 9) {

			number.of.genes.stocker <- append(number.of.genes.stocker,length(in.pathway))
			kegg.name.stocker <- append( kegg.name.stocker, unique.kegg.pathway.list[i])
			pca.input <- stock.list[in.pathway,4:(number.of.samples+3)]
					
      		        pca.input.numeric <- matrix(0,nrow=dim(pca.input)[1], ncol=number.of.samples)
						
			#the input as numeric
			for(j in 1: number.of.samples){
				pca.input.numeric[,j] <- as.numeric(pca.input[,j])
			}
    
			pca.input <- pca.input.numeric
                        pca.result <- prcomp(t(pca.input))
			t.pca.output.stocker <- t(pca.result$x[,c(1,2)])
			final.pca.output.stocker.PC1 <- rbind( final.pca.output.stocker.PC1, t.pca.output.stocker[1,] )
			final.pca.output.stocker.PC2 <- rbind( final.pca.output.stocker.PC2, t.pca.output.stocker[2,] )
	               
                        rand.str = gsub('0.', '', as.character(runif(1))) 
                        tempfile <- paste(loading.dir, '/loading', rand.str, sep='') 
                        loadings.images <- rbind(loadings.images, basename(tempfile))
                        loading(pca.result, tempfile)
		}
        }
   
	#bind the number of genes to the stockers of PC1 and PC2
	final.pca.output.stocker.PC1 <- cbind(number.of.genes.stocker,final.pca.output.stocker.PC1)
	final.pca.output.stocker.PC2 <- cbind(number.of.genes.stocker,final.pca.output.stocker.PC2)
 
	rownames(final.pca.output.stocker.PC1) <- kegg.name.stocker
	rownames(final.pca.output.stocker.PC2) <- kegg.name.stocker
        colnames(final.pca.output.stocker.PC1) <- c(colnames(stock.list)[3], paste("PC1:", colnames(stock.list)[4:(number.of.samples+3)], sep=""))
	colnames(final.pca.output.stocker.PC2) <- paste("PC2:", colnames(stock.list)[3:(number.of.samples+3)], sep="") 
        colnames(loadings.images) = 'Image'

        final <- cbind(final.pca.output.stocker.PC1, final.pca.output.stocker.PC2[,-1], loadings.images)        

	write.table(final, output.file.name.PC1, sep="\t", row.names=T, quote=F)
	#write.table(final.pca.output.stocker.PC2,output.file.name.PC2,sep="\t",row.names=T,quote=F)
}

#call
#pca.analysis("kegg.final.gse1786.txt" ,"PC1.yedek1.txt","PC2.yedek1.txt",24)
pca.analysis(m, outfile, outfile, num.samples)