R/RandomF.R

In statTarget: Statistical Analysis of Molecular Profiles

### Random forest function.
### nvarRF shows the number of variables in 
### Gini plot of Randomforest model (=< 100). 
### file The data of metabolites expression
### Multi result was outputed, such as MDS plot, 
### variable Gini plot and list results
### RandomF(file, nvarRF)
### The MDSplot and data matrix
RandomF <- function(file,nvarRF) {
  pwdfile = paste(getwd(), "/Univariate/DataTable.csv", sep = "")
  file = pwdfile
  x <- read.csv(file, sep = ",", header = TRUE)
  x.x = x[, 3:ncol(x)]
  rownames(x.x) = x[, 2]
  k = matrix(x[, 1], ncol = 1)
  slink = paste(getwd(), "/PreTable","/slink.csv", sep="")
  slink = read.csv(slink, header=TRUE)
  x.n = cbind(k, x.x)
  sorted = x.n[order(x.n[, 1]), ]
  g = c()
  for (i in 1:nrow(sorted)) {
    if (any(g == sorted[i, 1])) {
      g = g
    }
    else {
      g = matrix(c(g, sorted[i, 1]), ncol = 1)
    }
  }
  NoF = nrow(g)
  dirout.w = paste(getwd(), "/Univariate/RForest", sep = "")
  dirout.RF = paste(getwd(), "/Univariate/RForest/RF_Plots/", sep="")
  dir.create(dirout.w)
  dir.create(dirout.RF)
  
  for (i in 1:NoF) {
    for (j in 1:NoF) {
      if (i < j) {
        ni = paste("r.", ExcName(i,slink), ".csv", sep = "")
        nj = paste("r.", ExcName(j,slink), ".csv", sep = "")
        pwdi = paste(getwd(), "/Univariate/Groups/", ni, sep = "")
        pwdj = paste(getwd(), "/Univariate/Groups/", nj, sep = "")
        I = read.csv(pwdi, header = TRUE)
        J = read.csv(pwdj, header = TRUE)
        I = I[, -1]
        J = J[, -1]
        Ilf = matrix(rep(i),nrow(I))
        Jlf = matrix(rep(j),nrow(J))    
        colnames(Ilf) = c("lf")
        colnames(Jlf) = c("lf")
        I = cbind(Ilf,I)
        J = cbind(Jlf,J)
        IJ = rbind(I,J)
        #fin = ncol(sorted) - 1
        #we <- matrix(rep(NA, fin))
        #odds.radio...........................................................
        IJM <- as.matrix(IJ[,2:ncol(IJ)])
        outf <-as.factor(IJ[,1])
         
        #data(iris)
        #y <- as.factor(ifelse(iris$Species == "setosa" |
        #                        iris$Species == "virginica", 1, 0) )
        #xdata <- iris[,1:4]
        #yf <- c(1,1,2,2)
        #y <- as.factor(yf)
        message(paste("\n*Group.", ExcName(i,slink), sep = "")," Vs.", 
                paste(" Group.", ExcName(j,slink), sep = ""))
        nrep = 20
        mods <- rlply(nrep, randomForest(IJM, outf, ntree=500),
                      .progress = "text")
        dat <- matrix(rep(NA),nrep,length(IJM[1,]))
        #dat1 <- matrix(rep(NA),1,)
        dat1 <-c()
        for (k in 1:nrep) {
          for (h in 1:length(IJM[1,])){
            p <- function(x,y) { dat = as.list(importance(mods[[x]]))[[y]]}
            dat[k,h] <- p(k,h)
            dat1[k] <- median(mods[[k]][[4]][,1])
          }
          colnames(dat) = colnames(IJM)
          dat1 = matrix(dat1)
          colnames(dat1) = c("error.rate")
        }
        #write.csv(dat,"Gini_RF.csv")
        #write.csv(dat1,"error.rate.csv")
        Gini.ij = paste("Gini_", ExcName(i,slink), "vs", 
                        ExcName(j,slink), ".csv", sep = "")
        assign(Gini.ij, dat)
        write.csv(dat, paste(dirout.w, Gini.ij, sep = "/"))
        error.rate.ij = paste("error.rate_", ExcName(i,slink), "vs", 
                              ExcName(j,slink), ".csv", sep = "")
        assign(error.rate.ij, dat1)
        write.csv(dat1, paste(dirout.w, error.rate.ij, sep = "/"))
        
        ########### Randomforest MDS,boxplot ######
        
        rfmods <- randomForest(IJM, outf, ntree=200,proximity=TRUE)
        MDS = paste(dirout.RF, "MDSPlot_", ExcName(i,slink), "vs", 
                    ExcName(j,slink), ".pdf", sep="")
        pdf(MDS)
        rfplot <- MDSplot(rfmods, outf,k=2,
        #bg=rainbow(length(levels(as.factor(outf))))[unclass=outf], 
        #palette=c(1,2),
        pch=19, 
        #col=c("#D55E00","#009E73"), 
        palette=c("#D55E00","#009E73"), 
        xlab="Dimension 1", 
        ylab="Dimension 2", 
        cex=1.2, main=paste("MDS Plot ", ExcName(i,slink), " vs ", 
                            ExcName(j,slink), sep=""))
        legend("bottomright",
          legend=levels(outf),
          pch=19, 
          col=c("#D55E00","#009E73"), 
          cex=0.6)
        #text(Dim1, Dim2, rownames(euro.mds), cex=0.8, col="red")
        dev.off()
        RFscore.ij = paste("RFscore_", ExcName(i,slink), "vs", 
                           ExcName(j,slink), ".csv", sep = "")
        assign(RFscore.ij, rfplot$points)
        write.csv(rfplot$points, paste(dirout.w, RFscore.ij, sep = "/"))
        Giniplot = paste(dirout.RF, "Giniplot_", ExcName(i,slink), "vs", 
                         ExcName(j,slink), ".pdf", sep="")
        pdf(Giniplot)
        orderindex <- apply(dat,2,median)
        datindex <- cbind(orderindex,t(dat))
        sort.dat <- datindex[order(datindex[,1], 
                                   na.last=NA, decreasing = TRUE),]
        sort.dat <- t(sort.dat[,-1]) 
        sort.range <- sort.dat[,1:nvarRF]
        sort.datFilter <- flipdim_sT(sort.range, 2)
        par(mar=c(3, 10, 3, 10) + 0.1,mgp=c(1.5,0.5,0))
        colfunc <- colorRampPalette(c("grey48", "#D55E00"))
        #colfunc <- colorRampPalette(colors = brewer.pal(9,"reds"))
        boxplot(sort.datFilter, 
                main = paste("Gini Plot ", 
                ExcName(i,slink), " vs ", ExcName(j,slink), sep=""),
                notch = FALSE, col =colfunc(nvarRF), 
                cex =0.1, las = 1, xlab= "Mean Decrease Gini",
                ylab="Variables", horizontal = TRUE,
                pars = list(boxwex = 0.6, staplewex = 0.3,
                outwex = 0.3),lwd=0.3,
                cex.lab=0.8, cex.axis =0.4,medlwd=0.3, xaxt='n',yaxt="n")
        axis(2,labels= c(colnames(sort.datFilter)), at = 1:nvarRF, 
        cex = 0.3, tck= -0.015,cex.axis=0.3,las=1)
        axis(1,cex = 0.5, font= 0.05, tck= -0.015,cex.axis=0.3)
        grid(NULL,NA, lwd = 1,col = "gray")
        #axis(side=1, tcl=-0.1, labels=FALSE)
        dev.off()
        #library(plyr)
        #library(randomforest)
        #library(pracma)
      }
    }
  }
}