ribiosPlot (version 1.3.0)

cascadeOrder: Order rows of a matrix in the cascade order

Description

The 'cascade order' is defined by three criteria (1) Rows are divided into two groups by the condition given by 'dichotomy'. (2) The positive and negative rows are ordered respectively so that rows reaching its absolute maximal values in column n are ordered prior to rows reaching reaching its absolute maximal values in columns n+1, where n can be from 1 to column number minus one. (3) If two rows reach the maximum value at the same column, they are ordered by the (decreasing) order of the absolute value in that column.

Usage

cascadeOrder(matrix, dichotomy = c("maxabs", "mean", "median"))

Value

An integer vector of row indices in cascade order.

Arguments

matrix

A numeric matrix

dichotomy

How are the rows divided into two? By maximal abs value (default), mean value, or the median value of each row.

Details

See example for illustration of the idea.

Examples

checkBoard <- function(seed=1887) {
  set.seed(seed)
  mat <- matrix(rnorm(76, sd=1), ncol=4)
  delta <- 3
  for(i in seq(1, 16, 2)) {
    rowInd <- i:(i+1)
    colInd <- (i %/% 2) %% 4 +1
    delta <- ifelse(i>8, -6, 6) * c(0.6, 1)
    mat[rowInd, colInd] <- mat[rowInd, colInd] + delta
  }
  mat[17,1:4] <- rep(-1, 4)
  mat[18,1] <- NA
  mat[18,2] <- mat[18, 2] -6
  mat[19,1:4] <- rep(NA,4)
  rord <- sample(1:nrow(mat), replace=FALSE)
  mat <- mat[rord,]
  rownames(mat) <- sprintf("Row%d", 1:nrow(mat))
  return(mat)
}
myMat <- checkBoard(1887)
biosHeatmap(myMat, Rowv=FALSE, Colv=FALSE, dendrogram="none",
            zlim=c(-4,4), col="royalbluered",
            main="Original matrix")
## since dist by default does not accept rows full of NAs, we remove them in the example below
biosHeatmap(myMat[apply(myMat, 1, function(x) !all(is.na(x))),],
            Rowv=TRUE, Colv=TRUE, dendrogram="both",
            zlim=c(-4,4), col="royalbluered",
            main="hclust/dist clustering")
## note that cascadeOrder handles invariant rows and rows full of NA values
biosHeatmap(myMat[cascadeOrder(myMat),], Rowv=FALSE, Colv=FALSE, dendrogram="none",
            zlim=c(-4,4), col="royalbluered",
            main="Cascade order")