heatmap {apcluster}

Plot Heatmap

Description

Functions for Plotting of Heatmap

Usage

## S4 method for signature 'ExClust,missing'
heatmap(x, y, ...)
## S4 method for signature 'ExClust,matrix'
heatmap(x, y, ...)
## S4 method for signature 'ExClust,Matrix'
heatmap(x, y, ...)
## S4 method for signature 'ExClust,sparseMatrix'
heatmap(x, y, ...)
## S4 method for signature 'AggExResult,missing'
heatmap(x, y, ...)
## S4 method for signature 'AggExResult,matrix'
heatmap(x, y, Rowv=TRUE, Colv=TRUE,
    sideColors=NULL, col=heat.colors(12),
    base=0.05,  add.expr, margins=c(5, 5, 2),
    cexRow=max(min(35 / nrow(y), 1), 0.1),
    cexCol=max(min(35 / ncol(y), 1), 0.1), main=NULL, dendScale=1,
    barScale=1, legend=c("none", "col"), ...)
## S4 method for signature 'matrix,missing'
heatmap(x, y, ...)
## S4 method for signature 'missing,matrix'
heatmap(x, y, ...)

Arguments

x

a clustering result object of class APResult, ExClust, or AggExResult; for compatibility, x may also be a similarity matrix (see details below).

y

a similarity matrix

sideColors

character vector of colors to be used for plotting color bars that visualize clusters of the finest clustering level in x. This is done in a fashion similar to using RowSideColors or ColSideColors in the standard heatmap function. However, color bars are plotted either on both sides or not at all. The sideColors argument determines the coloring of both horizontal and vertical bars. If sideColors is shorter than the number of clusters in the finest clustering level, sideColors is recycled. In any case, a minimum number of two colors (two elements of sideColors) is required. If NA, no color bars are plotted. If NULL (default), color bars are only plotted if the finest cluster level does not only consist of single samples. In this case, the rainbow function is used to compute the vector of colors which is shuffled such that dissimilar colors are placed next to each other in the color bar.

col

color ramp used for the heatmap image; see image

Rowv

determines whether or not a row dendrogram should be plotted. If FALSE or NA, no row dendrogram is plotted. In any other case, a row dendrogram is plotted unless the number of clusters in the finest clustering level is less than 2. Note that, in the latter case, the actual values in Rowv are ignored, so this argument cannot be used to supply a previously computed dendrogram or re-ordering of elements as in the standard heatmap function.

Colv

determines whether or not a column dendrogram should be plotted. Fully analogous to Rowv, except that column dendrograms are never plotted if the similarity matrix y is non-quadratic.

base

fraction of height used for the very first join in dendrograms; see coerce-methods.

add.expr, margins, cexRow, cexCol, main

largely analogous to the standard heatmap function; to omit row/column labeling, set cexRow/cexCol to 0. The default for margins is a vector of length 3, where the third element is the right-hand side margin for the color legend (see legend argument). It remains unused (and can also be omitted) if no color legend is plotted.

dendScale

factor scaling the width of vertical and height of horizontal dendrograms; values have to be larger than 0 and no larger than 2. The default is 1 which corresponds to the same size as the dendrograms plot by the standard heatmap function

barScale

factor scaling the width of color bars; values have to be larger than 0 and no larger than 4. The default is 1 which corresponds to half the width of the color bars plot by the standard heatmap function

legend

if "col", then a color legend similar to filled.contour is added on the right-hand side of the heatmap plot; if "none" (default), no such legend is added.

...

see details below

Details

The heatmap functions provide plotting of heatmaps from several different types of input object. The implementation is similar to the standard graphics function heatmap. Plotting heatmaps via the plot command as available in previous versions of this package is still available for backward compatibility.

If heatmap is called for objects of classes APResult or ExClust, a heatmap of the similarity matrix in slot sim of the parameter x is created with clusters grouped together and highlighted in different colors. The order of clusters is determined by running aggExCluster on the clustering result x. This variant of heatmap returns an invisible AggExResult object.

If heatmap is called for an AggExResult object that contains all levels of clustering, the heatmap is displayed with the corresponding clustering dendrogram. If the AggExResult object is the result of running aggExCluster on a prior clustering result, the same heatmap plot is produced as if heatmap had been called on this prior clustering result, however, returning the cluster hierarchy's dendrogram. In the latter case, color bars are plotted to visualize the prior clustering result (see description of argument sideColors above).

All variants described above only work if the input object x contains a slot sim with the similarity matrix (which is only the case if the preceding clustering method has been called with includeSim=TRUE). In case the slot sim of x does not contain the similarity matrix, the similarity matrix must be supplied as second argument y.

All variants described above internally use heatmap with signature AggExResult,matrix, so all arguments list above can be used for all variants, as they are passed through using the ... argument. All other arguments, analogously to the standard heatmap function, are passed on to the standard function image. This is particularly useful for using alternative color schemes via the col argument.

The two variants with one of the two arguments being a matrix and one being missing are just wrappers around the standard heatmap function with the aim to provide compatibility with this standard case.

Value

see details above

Note

Similarity matrices can be supplied in dense or sparse format. Note, however, that sparse matrices are converted to full dense matrices before plotting heatmaps which may lead to memory and/or performance bottlenecks for larger data sets.

Author(s)

Ulrich Bodenhofer, Andreas Kothmeier, and Johannes Palme

References

https://github.com/UBod/apcluster

Bodenhofer, U., Kothmeier, A., and Hochreiter, S. (2011) APCluster: an R package for affinity propagation clustering. Bioinformatics 27, 2463-2464. DOI: doi:10.1093/bioinformatics/btr406.

See Also

APResult, AggExResult, ExClust, apcluster, apclusterL, aggExCluster, cutree-methods, plot-methods

Examples

## create two Gaussian clouds
cl1 <- cbind(rnorm(50, 0.2, 0.05), rnorm(50, 0.8, 0.06))
cl2 <- cbind(rnorm(50, 0.7, 0.08), rnorm(50, 0.3, 0.05))
x <- rbind(cl1, cl2)

## run affinity propagation using negative squared Euclidean
apres <- apcluster(negDistMat(r=2), x, p=-0.1)

## plot heatmap clustering run
heatmap(apres)

## rerun affinity propagation
## reuse similarity matrix from previous run
apres2 <- apcluster(s=apres@sim, q=0.6)

## plot heatmap of second run
heatmap(apres2, apres@sim)

## with alternate heatmap coloring, alternating color bars, and no dendrograms
heatmap(apres2, apres@sim, Rowv=NA, Colv=NA,
        sideColors=c("darkgreen", "yellowgreen"), col=terrain.colors(12))

## perform agglomerative clustering of affinity propagation clusters
aggres1 <- aggExCluster(apres@sim, apres2)

## plot heatmap
heatmap(cutree(aggres1, 2), apres@sim)

## perform agglomerative clustering of whole data set
aggres2 <- aggExCluster(negDistMat(r=2), x)

## show heatmap along with dendrogram
heatmap(aggres2)