###################################################################################
## Supplementary data                                                          ####
##      The R script reproducing ANOVA and PCA by using data deposited in GEO  ####
###################################################################################

# Download the three categories of required data from 
#   http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE30668.

# 1. parameter (in the Supplementary file list of the top page)
# 2. normalized cell data (found as a Supplementary file in each chip data's page)
# 3. expression levels (can be obtained by clicking the View full table button in each chip data's page). 

# The data categories should be kept separated by storing the data in different directories.
 #  The calculations will take place in the three directories one by one. 

#####
# 1 #  start the R and locate the working directory to that include "GSE30668_chip_parameter_SN.txt"
#####

      parameters<- read.table(file="GSE30668_chip_parameter_SN.txt", sep="\t", header=T)
      params <- data.matrix(parameters)[,2:13]
      save(params, file="parameters.RData")

##### 
# 2 # change the working directory to that include all the normalized probe-level PM data, such as "GSM760646.txt". 
#####
      nfiles<- list.files(pattern=".txt$")
      NumChip <- length(nfiles)
      ListData<-colnames(params)   
      groups <- sort(rep(1:3, 4)) # for example,
      groups<-as.factor(groups)   ; NumGroups <- length(levels(groups))

  save.image("parameters.RData")

# reading  matrix data from the normalized PM data files

   for (chips in 1:NumChip){
      nameofdata <- strsplit(nfiles[chips], ".txt")[[1]][1]
      nameoftxt <-nfiles[chips]
      DataTobeRead <- read.table(file=nameoftxt, sep="\t", na.strings=" ", fill=T, row.names=1, header=F)
         GeneContents <- rownames(DataTobeRead)
         NumContents <- length(GeneContents)
         b<-dim( DataTobeRead); a<-unlist( DataTobeRead); dim(a)<-b; rownames(a)<-GeneContents
         DataTobeRead<-a ; rm(a, b, GeneContents) # removing the list structure of the table
      # discarding data below the lowere limit and saturating point

         lowerlimit <- params[1, chips]
         upperlimit <- params[3, chips]
         DataTobeRead[DataTobeRead < lowerlimit] <- NA
         DataTobeRead[DataTobeRead > upperlimit] <- NA

      assign ( nameofdata ,  DataTobeRead)    # storing the matrix under the name of the chip 
      }    # chips

 save.image("datasets.RData")

#######################################################
###                ANOVA calculation                ###
#######################################################

# objects
   NumCel <- dim(get(strsplit(nfiles[1], ".txt")[[1]] ))[2]
   ANOVAres <- array(NA, dim=c(NumContents, 2))
   colnames(ANOVAres) <- c("group", "interaction")
   rownames(ANOVAres) <- rownames(get(strsplit(nfiles[1], ".txt")[[1]] ))

# preparing the factors that explains data
   experiments<-inlinedata <- 1:(NumCel*NumChip)+NA

   for (chips in 1:NumChip){
      experiments[1:NumCel+(chips-1)*NumCel] <- groups[chips] }

   cells <- rep( 1:NumCel, NumChip)
   cells<-as.factor(cells); experiments<-as.factor(experiments) # specify the type of objects

# uptaking each gene data and ANOVA
   for (genes in 1:NumContents){
    for (chips in 1:NumChip){
    inlinedata[1:NumCel+(chips-1)*NumCel] <- get(strsplit(nfiles[chips], ".txt")[[1]] )[genes, ]} 
   ###### balancing the test by removing incomplete measurements ###############
      na_cells <-get(strsplit(nfiles[1], ".txt")[[1]] )[genes, ]
       for (chips in 2:NumChip){
          na_cells <- na_cells+get(strsplit(nfiles[chips], ".txt")[[1]] )[genes, ]} #uptake
      na_cells <- na_cells*0+1   # comprlete cells will become one
      Num_non_na_cells<-sum(na_cells, na.rm=T)
      balancing_filter <- rep(na_cells*0, NumChip)
      inlinedata <-inlinedata+ balancing_filter   # removing      

   if (Num_non_na_cells > 1){
      #### end of balancing ###########

      AOVres <- summary(aov( inlinedata ~ cells+experiments+cells:experiments))
      ANOVAres[genes,"group"] <- AOVres[[1]]["experiments","Pr(>F)" ]
      ANOVAres[genes,"interaction"] <-AOVres[[1]]["cells:experiments","Pr(>F)" ]
   }   }  # if (Num_non_na_cells   and  for (genes

hist(ANOVAres[,"group"], breaks=c(0:100/101,1))
save(ANOVAres, file="ANOVAres.rdata")

#####
# 3 #  Set the working directory on the same one where the expression levels are located.
#####
    # The files could be obtained by clicking "view full table" button in the GEO page
    # and stored as a text file, such as "GSM760657.acc.cgi.txt"

# master data
    NumChip<-12
   list_p_files<- list.files(pattern=".txt$")
     mastermatrix <-array(NA, dim=c( dim(ANOVAres)[1], NumChip))
      for (chip in 1:NumChip)   {
         data_read<-read.table(file=list_p_files[chip], header = T, sep = "\t", skip=4, fill=T)
          mastermatrix[ ,chip] <- data_read[1:31099,4]
                }

   colnames(mastermatrix) <- list_p_files
   gn<- rownames(mastermatrix) <- rownames(data_read)[1:31099]
   controls<-which(substring(gn, 1, 4)=="AFFX")
   row.names(mastermatrix) <- gn

 p_values <- ANOVAres[,"group"]

 groups<- c("T", "A", "C")

 representative<- cbind( apply(mastermatrix[, 1:4], 1, mean, na.rm=T), apply(mastermatrix[, 1:4+4], 1, mean, na.rm=T), apply(mastermatrix[, 1:4+8], 1, mean, na.rm=T))
	colnames(representative)<-groups


#######################################################
###                   PCA                           ###
#######################################################
# The pc's and the loadings are estimated on the correlation matrixes among representative data of selected genes.
 
# selecting genes 
   # to those positive in ANOVA

   selected_genes<-which(ANOVAres[,"group"] < 0.001/2) 
   data_subjected <- t(representative)*0
   data_subjected[, selected_genes] <- t(representative)[, selected_genes]

   means<-data_subjected[2, ]
   data_subjected<-sweep(data_subjected, 2, means) 	

   data_subjected[, controls] <-0 

 svdres<-svd(data_subjected)  # SVD

  data_master<-t(mastermatrix)*0
  data_master[, selected_genes] <-t(mastermatrix)[, selected_genes]
  data_master<-sweep(data_master, 2, means)
  data_master[, controls]<- 0

 pcs <- data_master %*% svdres$v 
  pcs<-pcs/sqrt(length(selected_genes))

 loadings <- t(data_subjected)%*% svdres$u
  loadings <- loadings /sqrt(3)

#######################################################
###                   output                        ###
#######################################################

plot(loadings, col="gray50", xlab="PC1", ylab="PC2")
     label <- c("T","T","T","T","A","A","A","A","C","C","C","C")

   for(i in 1:length(label)){
      text(labels=label[[i]][1], x=pcs[i, 1], y=pcs[i, 2], col="Blue4") }

  abline(h=0,  lty=1, col="White"); abline(v=0,  lty=1, col="White"); abline(h=0, lty=3); abline(v=0, lty=3)
  abline(h=-.1, lty=3); abline(h=.1, lty=3); abline(v=-.16, lty=3); abline(v=.16, lty=3)

#######################################################
###                  information                    ###
#######################################################

# <title> ANOVA to PCA </title>
# <subtitle> Supplement Data </subtitle>
# <version> R2.4.1 & R 2.11.1 </version>
# <date> 19 July 2011 </date>
# <author> Tomokazu Konishi </author>
# <EOF>