R/see.genes.R

In maSigPro: Significant Gene Expression Profile Differences in Time Course Gene Expression Data

"see.genes" <-
function (data, edesign = data$edesign, time.col = 1, repl.col = 2, 
    group.cols = c(3:ncol(edesign)), names.groups = colnames(edesign)[3:ncol(edesign)], 
    cluster.data = 1, groups.vector = data$groups.vector, k = 9, k.mclust=FALSE,   
    cluster.method = "hclust", distance = "cor", agglo.method = "ward.D", 
    show.fit = FALSE, dis = NULL, step.method = "backward", min.obs = 3, 
    alfa = 0.05, nvar.correction = FALSE, show.lines = TRUE, iter.max = 500, 
    summary.mode = "median", color.mode = "rainbow", cexlab = 1, legend = TRUE, 
    newX11 = TRUE,  ylim = NULL, main = NULL, item ="genes",...) 
{

#--------------------------------- DATA PREPARATION ---------------------------------

    time = edesign[, time.col]
    repvect = edesign[, repl.col]
    groups = edesign[, group.cols]
    narrays <- length(time)
    if (!is.null(dim(data))) {
        dat <- as.data.frame(data)
        clusterdata <- data
    }
    else {
        clusterdata <- data[[cluster.data]]
        dat <- as.data.frame(data$sig.profiles)
    }

if (nrow(dat) > 1) {

#--------------------------------- NAs REMOVEMENT ---------------------------------
        dat <- as.data.frame(dat[, (ncol(dat) - length(time) + 1):ncol(dat)])
        count.noNa <- function(x) (length(x) - length(x[is.na(x)]))
        dat <- dat[which(apply(as.matrix(dat), 1, count.noNa) >=  length(unique(repvect))), ]
        clusterdata <- dat

#--------------------------------- NAs TREATMENT ---------------------------------
# If cluster.data is beta or t.values, they are changed for 0s.
# With profiles, for mean value of the same type of replicate.

  if (any(is.na(clusterdata))) {
	if( cluster.method == "kmeans" || cluster.method=="Mclust") {

	  if ( all(cluster.data != 1, cluster.data != "sig.profiles") ) {
			 clusterdata[is.na(clusterdata)] <- 0
        }
	  else {
	  mean.replic <- function(x){ tapply(as.numeric(x), repvect, mean,na.rm=TRUE) }
	  MR <- t( apply(clusterdata, 1, mean.replic) )
	  # por si acaso todos los valores de una misma réplica son NAs:
		 if( any(is.na(MR)) ) {
		   row.mean <- t(apply(MR, 1, mean, na.rm=TRUE ))
		   MRR<-matrix(row.mean,nrow(MR),ncol(MR) )
		   MR[is.na(MR)] <- MRR[is.na(MR)]
		  }
	  data.noNA <- matrix(NA,nrow(clusterdata),ncol(clusterdata))
	  u.repvect <- unique(repvect)
	  for (i in 1:nrow(clusterdata)){
	    for(j in 1:length(u.repvect)) {
	    data.noNA[i, repvect==u.repvect[j] ] = MR[i,u.repvect[j]]
	  }
	  }
	clusterdata<-data.noNA
	}
  }
  }
        
#--------------------------------------------------------------------------
# CLUSTERING
#--------------------------------------------------------------------------
        if (!is.null(clusterdata)) {
# heatmap(as.matrix(clusterdata))
             k <- min(k, nrow(dat), na.rm = TRUE)

#--------------------------------------------
# hclust
#--------------------------------------------
            if (cluster.method == "hclust") {
                if (distance == "cor") {
                  dcorrel <- matrix(rep(1, nrow(clusterdata)^2), nrow(clusterdata), nrow(clusterdata)) - cor(t(clusterdata), 
                    use = "pairwise.complete.obs")
                  clust <- hclust(as.dist(dcorrel), method = agglo.method)
                  c.algo.used = paste(cluster.method, "cor", agglo.method, sep = "_")
                }
                else {
                  clust <- hclust(dist(clusterdata, method = distance), method = agglo.method)
                  c.algo.used = paste(cluster.method, distance, 
                    agglo.method, sep = "_")
                }
                cut <- cutree(clust, k = k)
            }
#--------------------------------------------
# kmeans
#--------------------------------------------
            else if (cluster.method == "kmeans") {
                cut <- kmeans(clusterdata, k, iter.max)$cluster
                c.algo.used = paste("kmeans", k, iter.max, sep = "_")
            }

#--------------------------------------------
# Mclust
#--------------------------------------------
   else if (cluster.method == "Mclust") {
	if (k.mclust) {
	  my.mclust <- Mclust(clusterdata)
	  k=my.mclust$G
	}
	else {	
	my.mclust <- Mclust(clusterdata,k) 
	}
	cut <- my.mclust$class
	c.algo.used = paste("Mclust", k, sep = "_")
	}

#--------------------------------------------------------------------------
            else stop("Invalid cluster algorithm")

#--------------------------------------------------------------------------
            if (newX11) 
                X11()
            groups <- as.matrix(groups)
            colnames(groups) <- names.groups
            if (k <= 4) 
                par(mfrow = c(2, 2))
            else if (k <= 6) 
                par(mfrow = c(3, 2))
            else if (k > 6) 
                par(mfrow = c(3, 3))
            for (i in 1:(k)) {
                PlotProfiles(data = dat[cut == i, ], repvect = repvect, 
                  main = i, ylim = ylim, color.mode = color.mode, 
                  cond = rownames(edesign), item = item, ...)
            }
            if (newX11) 
                X11()
            if (k <= 4) {
                par(mfrow = c(2, 2))
                cexlab = 0.6
            }
            else if (k <= 6) {
                par(mfrow = c(3, 2))
                cexlab = 0.6
            }
            else if (k > 6) {
                par(mfrow = c(3, 3))
                cexlab = 0.35
            }
            for (j in 1:(k)) {
                PlotGroups(data = dat[cut == j, ], show.fit = show.fit, 
                  dis = dis, step.method = step.method, min.obs = min.obs, 
                  alfa = alfa, nvar.correction = nvar.correction, show.lines = show.lines, time = time, 
                  groups = groups, repvect = repvect, summary.mode = summary.mode, 
                  xlab = "time", main = paste("Cluster", j, sep = " "), 
                  ylim = ylim, cexlab = cexlab, legend = legend, 
                  groups.vector = groups.vector, item = item, ...)
            }
        }
        else {
            print("warning: impossible to compute hierarchical clustering")
            c.algo.used <- NULL
            cut <- 1
        }
    }
  else if (nrow(dat) == 1) {
        if (newX11) 
            X11()
        PlotProfiles(data = dat, repvect = repvect, main = NULL, 
            ylim = ylim, color.mode = color.mode, cond = rownames(edesign), 
            ...)
        if (newX11) 
            X11()
        PlotGroups(data = dat, show.fit = show.fit, dis = dis, 
            step.method = step.method, min.obs = min.obs, alfa = alfa, nvar.correction = nvar.correction,
            show.lines = show.lines, time = time, groups = groups, 
            repvect = repvect, summary.mode = summary.mode, xlab = "time", 
            main = main, ylim = ylim, cexlab = cexlab, legend = legend, 
            groups.vector = groups.vector, ...)
        c.algo.used <- NULL
        cut <- 1
    }
    else {
        print("warning: NULL data. No visualization possible")
        c.algo.used <- NULL
        cut <- NULL
    }
    OUTPUT <- list(cut, c.algo.used, groups)
    names(OUTPUT) <- c("cut", "cluster.algorithm.used", "groups")
    OUTPUT
}