#******************************************************************************************
#Compound Context Analysis
#[Author]Tara A. Gianoulis
#[Last Modified] September 3, 2011
#******************************************************************************************

#Table of contents
#Normalization
#Permutation Testing
#Accessory Methods
#Compound Co-expression and Co-occurrence Functions
#Plotting and Printing

#******************************************************************************************
#Functions for Normalization
#******************************************************************************************

#----Quantile Normalize---#
#Quantile Normalize
#(1) Sort columns of x
#(2) Compute row medians
#(3) Use approx function to do the linear interpolation

quantileNormalize<-function (x)
{
  xm <- apply(x, 2, sort)
  xxm <- apply(xm,1, median)
  xr <- c(apply(x, 2, rank))
  q<-array(approx(1:nrow(x), xxm, xr)$y, dim(x), dimnames(x))
  return (q)
}

#---Median Normalize---#
#if(x1 == med) medNorm=1
#if(x1 == 0) medNorm = 0

medianNormalize<-function(X){
        #Compute median of each column
        m<-apply(X,2,median)
        
        #Create median matrix by repeating elements for as many rows as in X
        xRows<-nrow(X)
        xCols<-ncol(X)
        M<-t(matrix(m,xRows,xCols))

        #Subtract each element in X by column median
        medNorm<-X-M

        return(medNorm)
}

#******************************************************************************************
#Functions for Shuffling
#******************************************************************************************

#REQUIRES createRandIndex
createShuffledMatrix<-function(X){
        M<-apply(X,2,createRandIndex)
        return(M)
}       


#Compute shuffled versions of each column
createRandIndex<-function(X){
        randIndex<-sample(X)
        return (X[randIndex])
}

#This requires rfindCor from RNASeqfunctions.R
#Create distribution of random correlations
computeRandomCor<-function(A, B,times){

  shuffledVal<-0
  for(i in 1:times){
    shuffledA<-matrix(sample(A), nrow(A), ncol(A) )
    shuffledCor<-rfindCor(shuffledA, B)
    shuffledVal<-c(shuffledCor, shuffledVal)
  }
  return (shuffledVal)
}


#******************************************************************************************
#Accessory Methods
#******************************************************************************************

#Returns only the t-test from the ttest object
getTfromStruct<-function(struct){
        p<-unlist(lapply(struct, "[",1));
        return(p)
}

#Returns only the pvalues from the ttest object
getPvalfromStruct<-function(struct){
        p<-unlist(lapply(struct, "[",3));
        return(p)
}

#Compute fold change
computeFoldChange<-function(M){
        M<as.matrix(M);
        dimM<-dim(M);
        M<-M+1;
        foldChange<-matrix(nrow=dimM[1], ncol=dimM[2]*dimM[2]);
        colCount<-1;
        for(i in 1:dimM[2]){
                for(j in 1:dimM[2]){
                        foldChange[,colCount]<- log2(M[,i]/M[,j])
                        colCount<-colCount+1;
                }
        }
        return (foldChange);
}


#******************************************************************************************
#Compound Co-expression and Co-occurrence Functions
#******************************************************************************************

#---Compound Co-expression Score---#
#Requires corrCmpdProfileR and fastercorrectedPval
computeCCE<-function(A,B, divisions,last, shuffleLength,shuffledVal){

			  maxM<-corrCmpdProfileR(A, B)

			  #Compute corrected pval

			  p<-matrix(0, nrow=nrow(maxM), ncol=divisions)
			  begin<-1
			  incrementor<-last

			  for(i in 1:divisions){

 			   	p[,i]<-fastercorrectedPval(shuffledVal[begin:last], maxM )
 			   	begin <-begin + incrementor
  			   	last <- last + incrementor
			   }
			   pval<-rowSums(p)/shuffleLength

			   maxMC<-cbind(maxM, pval)
			   return (maxMC)
}


#This requires rfindCor 
corrCmpdProfileR<-function(A,B){
#Correlate all expression profiles with the set of compound profiles
  X<-rfindCor(A,B)
#Order absolute value of correlation indices, ties appear to be random...
  orderedIndex<-apply(abs(X), 1,order, decreasing=TRUE)

#Select the maximum abs correlation and add the row number of the compound profile to the gene id matrix resulting in a 3 col vector geneNumber, profileId, cor
relation

  maxR<-cbind(1:nrow(X),orderedIndex[1,], X[cbind(1:nrow(X), orderedIndex[1,])])
  return (maxR)
}


#Computes correlation of matrix vs matrix and returns r2
rfindCor <-function(X,Y){
        rsquared<-matrix(0, nrow=nrow(X), ncol=nrow(Y) )
        for(i in 1:nrow(X)){
              for(j in 1:nrow(Y)){
                    temp<-cor.test(X[i,],Y[j,], method="pearson")
                    rsquared[i,j]=temp$estimate;
              }
        }
        return  (rsquared)
}

#Just counts - leaves the rest for later
fastercorrectedPval<-function(shuffledVal, maxR){
  numerator<-matrix(0, nrow(maxR), 1)
  shuffledAb<-abs(shuffledVal)
  maxAb<-abs(maxR)
  for(i in 1:nrow(maxR)){
    numerator[i]<-length(which(shuffledAb>=maxAb[i,3]))
  }
  return(numerator)
}


#---Compound Co-occurrence Score---#
#From Luke's Associates Daylight
#Lance & Williams
#Bray-Curtis nonmetric coefficient
#DLW= (L-(Bi xnor Bj) )/(Bi +Bj)
#Requires xnor function
computefasterDLW<-function(data){
	L<-dim(data)[2]
	rowNum<-dim(data)[1]
	rs<-rowSums(data)
	DLW<-matrix(0,rowNum, rowNum)
	xnorMat<-matrix(rowSums(apply(x, 1, xnor, y=x)),rowNum, L)
	numerator<-L-xnorMat
	for( i in 1:rowNum) {
		for(j in 1:rowNum){
	   	      denom<-rs[i]+rs[j]
	  	      numerator<- L-xnorMat[i,j]
	  	      DLW[i,j]<-numerator/denom
     		 }
	}
	return ( as.dist(DLW) )
}

#Additional Logic Gates
nor<-function(x,y){!(x | y)}
xnor<-function(x,y){(x & y) | (!x & !y)}


#******************************************************************************************
#FUNCTIONS for Pretty Plotting 
#******************************************************************************************

#This function expects a color vector to be given alongside dataset
makePlotComplex<-function(X, color) {
       maxR<-max(X)
       minR<-min(X)
       plot(1:ncol(X), X[1, ], ylim=c(minR, maxR), col=color )
			   
	for(i in 1:nrow(X) ){
	      lines(1:ncol(X), X[i, ], ylim=c(minR, maxR), col=color )
	}
}

#******************************************************************************************
#FUNCTIONS: Highly specific functions, limited general utility
#******************************************************************************************

getCounts<-function(counts,lng){
	countType<-0
	for(i in 1: lng ){
	     ctype<-length( which ( counts==i) )
	     countType<c(countType, ctype)
	}
	return (countType)
}