###-------------------------------------------------------###
### The code below will help you import the data set into ###
###       R so further analysis can be done.              ###
###-------------------------------------------------------###

# First download the two data sets from
# http://www-genome.wi.mit.edu/cgi-bin/cancer/datasets.cgi
# Delete all columns (and rows) without any expression values.
# This should result in matrics consisting of only the expression
# values. Save the data sets as comma separated files (CSV) and name
# the train and independent data sets as train and indep, respectively
# on the C drive in a folder named temp.
 
	 
	tXs <- read.table(file="C:/temp/train.csv", 
						sep=",", header=F)
	iXs <- read.table(file="C:/temp/indep.csv", 
						sep=",", header=F)

# Arrange the data so that genes are the columns not rows.

	tXs <- as.matrix(t(tXs)) # Matrix is now 38 x 7129
	iXs <- as.matrix(t(iXs)) # Matrix is now 34 x 7129

	full <- rbind(tXs, iXs) # Matrix is 72 x 7129

	# Threshold values at 30
	
		full[full < 30] <- 30
		
	# Normalize the data 
		
		Xs.rank <- apply(full, 1, order)
		Xs.rank.var <- apply(Xs.rank, 1, var) # variance vector
		Invariant.gene <- (1:7129)[Xs.rank.var < median(Xs.rank.var)]
			# Above keeps only 50% invariant genes for baseline normalization
			
		logX <- log(full)
		xbar.vec <- apply(logX[,Invariant.gene], 1, FUN=mean)

	normalize <- function(x){
			(x - mean(x))/sqrt(var(x))
	}

		logXmat <- as.data.frame(logX - xbar.vec)
				logXmat <- apply(logXmat, 2, normalize)

		# Separate into the training and independent data sets
		
		train <- logXmat[1:38,]
		indep <- logXmat[39:72,]