clustComp (version 1.0.0)

drawTreeGraph: Plot the bi-graph determined by the branches in the tree and the flat clusters

Description

drawTreeGraph plots both a hierarchical tree, either complete or pruned, and a flat clustering, connected with edges whose thickness is proportional to the number of elements shared by branches and clusters, to form a weighted bi-graph. Its usage is mainly internal, as part of the visualisation of the hierarchical.look.ahead function. The size of each cluster is also displayed.

Usage

drawTreeGraph(weight, current.order, coordinates, tree, dot = TRUE,  line.wd = 3, main = NULL, expanded = FALSE, hclust.obj = NULL,  labels = NULL, cex.labels = 1)

Arguments

weight

a contingency matrix containing the intersection sizes (edge weights) between branches in the tree and clusters from the flat partitioning.

current.order

a list of two components; the first one is a vector with the branches (rows of the matrix weight) in the ordering in which they are drawn; the second one provides the ordering for flat clusters (columns of weight). Both are drawn from bottom upwards.

coordinates

a list of two components; the first one is a vector providing the Y-coordinates, from bottom upwards, for the branches, whereas the second one provides the Y-coordinates, from bottom upwards, for the flat clusters.

tree

a list with two components: $heights, a vector describing the heights at which the different branches in the tree are agglomerated, and $branches, a matrix of 3 columns; the i-th row contains as first element the branch split at the i-th allowed splitting, and as second and third elements, the corresponding children.

dot

a Boolean parameter; if TRUE then the last split in the children-tree is shown with a red open circle.

line.wd

a number indicating the width of the thickest edge(s) in the bigraph.

main

a character string for the plot title.

expanded

a Boolean parameter indicating whether the hierarchical tree should be plotted complete or with its branches collapsed.

hclust.obj

an hclust object describing the how the leaves are merged and the ordering of the branches, which might have been changed by the gravity-centre algorithm.

labels

a vector indicating the labels for the leaves in the expanded tree.

cex.labels

a number indicating the magnification used for the labels of the leaves.

Value

a list of components including:

b.coord

a vector indicating the Y coordinates of the nodes in the bi-graph representing the branches of the hierarchical tree.

f.coord

a vector indicating the Y coordinates of the nodes in the bi-graph representing the flat clusters.

x.coords

a vector of two components indicating the X coordinates at which each layer of the bi-graph is represented.

Details

The drawTreeGraph allows visualising the comparison of a hierarchical clustering, drawn on the left hand side of the plot, and a flat clustering, represented on the right hand side. The tree branches are labelled by their original labels preceded by 'B'; if the function is called as part of the flatVShier algorithm, then the standard notation for initial branch labels is that of hclust objects: if the label is a negative integer it corresponds to a leaf; if it is a positive integer, then it corresponds to a branch that agglomerates at least two elements, and the number represent the stage at which the branch was formed. The last splitting can be optionally highlighted with a red open circle upon the parent-node. The flat clusters are labelled by their original labels preceded by 'F'.

References

Torrente, A. et al. (2005). A new algorithm for comparing and visualizing relationships between hierarchical and flat gene expression data clusterings. Bioinformatics, 21 (21), 3993-3999.

See Also

flatVShier

Examples

    ### simulated data
    parent.clustering <- c(rep(1, 15), rep(2, 10))
    # replace the branch '2' by its children '3' and '4' 
    children.clustering <- c(rep(1, 15), rep(3, 5), rep(4, 5))
    flat.clustering <- c(rep(1, 6), rep(2, 6), rep(3, 4), rep(4, 9))
    split <- rbind(c(0, 1, 2), c(2, 3, 4))
    weight <- table(children.clustering, flat.clustering)
    current.order <- list(c(3, 4, 1), 1:4)
    coordinates <- list(c(-1, 0, 1), c(-1.5, -0.5, 0.5, 1.5))
    tree<-list(heights = c(1, 0.8), branches = split)
    drawTreeGraph(weight, current.order, coordinates, tree)

    ### expanded tree
    set.seed(0)
    myData <- matrix(rnorm(50), 10, 5)
    myData[1:5,] <- myData[1:5, ] + 2  # two groups
    flat.clustering <- kmeans(myData, 2)$cluster
    hierar.clustering <- hclust(dist(myData))
    weight <- matrix(c(5, 0, 0, 5), 2, 2)
    colnames(weight) <- 1:2; rownames(weight) <- c(6,8)
    current.order <- list(c(6, 8), 1:2)
    coordinates <- list(c(0.25, 0.75), c(0.25, 0.75))
    tree <- list(heights = hierar.clustering$height[9], 
        branches = matrix(c(9, 6, 8), 1, 3))
    drawTreeGraph(weight, current.order, coordinates, tree, 
        expanded = TRUE, hclust.obj = hierar.clustering, 
        dot = FALSE)