ChemmineR (version 2.24.2)

cluster.visualize: visualize clustering result using multi-dimensional scaling

Description

'cluster.visualize' takes clustering result returned by 'cmp.cluster' and generate multi-dimensional scaling plot for visualization purpose.

Usage

cluster.visualize(db, cls, size.cutoff, distmat=NULL, color.vector=NULL, non.interactive="", cluster.result=1, dimensions=2, quiet=FALSE, highlight.compounds=NULL, highlight.color=NULL, ...)

Arguments

db

The desciptor database, in the format returned by 'cmp.parse'.

cls

The clustering result returned by 'cmp.cluster'.

size.cutoff

The cutoff size for clusters considered in this visualization. Clusters of size smaller than the cutoff will not be considered.

distmat

A distance matrix that corresponds to the 'db'. If not provided, it will be computed on-the-fly in an efficient manner.

color.vector

Colors to be used in the plot. If the number of colors in the vector is not enough for the plot, colors will be reused. If not provided, color will be generated and randomly sampled from 'rainbow'.

non.interactive

If provided, will enable the non-interactive mode, and the plot will be in an eps file named after this value.

cluster.result

Used to select the clustering result if multiple clustering results are present in 'cls'.

dimensions

Dimensionality to be used in visualization. See details.

quiet

Whether to supress the progress bar.

highlight.compounds

A vector of compound IDs, corresponding to compounds to be highlighted in the plot. A highlighted compound is represented as a filled circle.

highlight.color

Color used for highlighted compounds. If not set, a highlighted compounds will have the same color as that used for other compounds in the same cluster.

...

Further arguments will be passed to 'cmp.similarity' to calculate similarity matrix.

Value

This function returns a data frame of MDS coordinates and clustering result. This value can be passed to 3D plot utilities such as 'scatterplot3d' and 'rggobi'.The last column of the output gives whether the compounds have been clicked in the interactive mode.

Details

'cluster.visualize' internally calls the 'cmdscale' function to generate a set of points in 2-D for the compounds in selected clusters. Note that for compounds in clusters smaller than the cutoff size, they will not be considered in this calculation - their entries in 'distmat' will be discarded if 'distmat' is provided, and distances involving them will not be computed if 'distmat' is not provided.

To determine the value for 'size.cutoff', you can use 'cluster.sizestat' to see the size distribution of clusters.

Because 'cmp.cluster' function allows you to perform multiple clustering processes simultaneously with different cutoff values, the 'cls' parameter may point to a data frame containing multiple clustering results. The user can use 'cluster.result' to specify which result to use. By default, this is set to 1, and the first clustering result will be used in visualization. Whatever the value is, in interactive mode (described below), all clustering result will be displayed when a compound is selected in the interactive plot.

If the colors provided in 'color.vector' are not enough to distinguish clusters by colors, the function will silently reuse the colors, resulting multiple clusters colored in the same color. We suggest you use 'cluster.sizestat' to see how many clusters will be selected using your 'size.cutoff', or simply provide no 'color.vector'.

If 'non.interative' is not set, the final plot is interactive. You will be able to select points by clicking them. When you click on any point, information about the compound represented by that point will be displayed. This includes the cluster ID, cluster size, compound index in the SDF and compound name if any. You can then perform another selection. To exit this process, right click on X11 device or press ESC in non-X11 device (Quartz and Windows).

By default, 'dimensions' is set to 2, and the built-in 'plot' function will be used for plotting. If you need to do 3-Dimensional plotting, set 'dimensions' to 3, and pass the returned value to 3D plot utilities, such as 'scatterplot3d' or 'rggobi'. This package does not perform 3D plot on its own.

See Also

cmp.parse, cmp.cluster, cluster.sizestat

Examples

## Load sample SD file
# data(sdfsample); sdfset <- sdfsample

## Generate atom pair descriptor database for searching
# apset <- sdf2ap(sdfset) 

## Loads same atom pair sample data set provided by library
data(apset) 
db <- apset

## cluster db with 2 cutoffs
clusters <- cmp.cluster(db, cutoff=c(0.5, 0.4))

## Return size stats
sizestat <- cluster.sizestat(clusters)

## Visualize results, using a cutoff of 3, write to file 'test.eps'
coord <- cluster.visualize(db, clusters, 2, non.interactive="test.eps")

## Not run: 
# ## visualize it in interactive mode, using a cutoff of 3 and the 2nd clustering result
# coord <- cluster.visualize(db, clusters, cluster.result=2, 3)
# 
# ## 3D visualization with scatterplot3d
# coord <- cluster.visualize(db, clusters, 3, dimensions=3)
# library(scatterplot3d)
# scatterplot3d(coord)
# 
# ## End(Not run)