score.crossing: Computation of the aesthetics-based score of the parent and...
In clustComp: Clustering Comparison Package

Description Usage Arguments Details Value Author(s) References See Also Examples

View source: R/score.crossing.R

score.crossing computes the value of the scoring function based on the aesthetics of the bi-graph formed when comparing a dendrogram and a flat clustering, for both the parent-tree and the children-tree; the children-tree consists of the same branches as the parent-tree, except for the parent node, that has been split and replaced by some of its descendants.

1	score.crossing(weight.1, weight.2, N.cross)

`weight.1`	a matrix of dimension `(m-1)xn` containing the intersection sizes (edge weights) between branches in the parent-tree and clusters from the flat partitioning. The ordering of the rows and columns is irrelevant for the computation of the score.
`weight.2`	a matrix of dimension `(m+k)xn` containing the intersection sizes (edge weights) between branches in the children-tree and clusters from the flat partitioning. `k` takes on values in 0,1,...,L, where L is the maximum number of steps that the comparison algorithm is allowed to look ahead. The ordering of the rows corresponding to branches that are not descendants of the parent node must coincide with that of the matrix `weight.1` after discarding the parent node. The ordering of the columns is irrelevant for the computation of the score.
`N.cross`	the number of edge crossings induced in the subtree formed by the descendants of the parent-node.

The decision to split a given parent-node is based on achieving a better score for the children-tree than for the parent-tree. In the case of score.crossing, a better score is reflected by a larger value of the scoring function, which rewards few thicker edges, penalises many smaller edges, and accounts for the number of edge crossings in the resulting bigraph. The descendants of the parent-node considered in the children-tree are its two children if no look-ahead is carried out; otherwise, the descendants will reach subsequent generations and their number will increase by one at each look-ahead step.

a list containing the following components:

`sc1`	the value of the scoring function for the parent-tree.
`sc2`	the value of the scoring function for the children-tree.

Aurora Torrente aurora@ebi.ac.uk and Alvis Brazma brazma@ebi.ac.uk

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.

score.it, dyn.cross, flatVShier

    ### simulated data
    parent.clustering <- c(rep("B1", 5), rep("B2", 10), rep("B3", 10))
    # replace the branch 'B2' by its children 'B4' and 'B5'
    children.clustering <- c(rep("B1", 5), rep("B4", 3), rep("B5", 7),
        rep("B3", 10))
    flat.clustering <- c(rep(1, 6), rep(2, 6), rep(3, 4), rep(4, 9))
    # the ordering of flat clusters is '1','2','3' and '4'.
    parent.weights <- table(parent.clustering, flat.clustering)
    children.weights <- table(children.clustering, flat.clustering)
    descendant.cross <- dyn.cross(children.weights[c('B4', 'B5'), ])
    score.crossing(parent.weights, children.weights, descendant.cross) 
    ## better score for the parent.tree