tight.clust {tightClust}

Tight Clustering

Description

This function could perform the tight clustering algorithm.

Usage

tight.clust(x, target, k.min, alpha = 0.1, beta = 0.6,
top.can = 7, seq.num = 2, resamp.num = 10,
samp.p = 0.7, nstart = 1, remain.p = 0.1,
k.stop = 5, standardize.gene=TRUE, random.seed=NULL)

Arguments

x

Input data, should be matrix. Each row represents a gene (object) to be clustered. Gene (object) names are usually represented in the row names and sample (feature) names are in the column names of the matrix.

target

The total number of clusters that the user aims to find.

k.min

The starting point of k0. See 'Details' for more information.

alpha

The threshold of comembership index. Default value is suggested to be used.

beta

The threshold of clusters stably found in consecutive k0. Default value is suggested to be used.

top.can

The number of top (size) candidate clusters for a specific k0. Default value is suggested to be used.

seq.num

The number of subsequent k0 that finds the tight cluster. Default value is suggested to be used.

resamp.num

Total number of resampling to obtain comembership matrix. Default value is suggested to be used.

samp.p

Percentage of subsamples selected. Default value is suggested to be used.

nstart

Number of different random inital for K-means. Default value is suggested to be used.

remain.p

Stop searching when the percentage of remaining points <= remain.p. Default value is suggested to be used.

k.stop

Stop decreasing k0 when k0<=k.stop. Default value is suggested to be used.

standardize.gene

It specifies whether to standardize each gene vector to mean 0 and sd 1. Default value is suggested to be used.

random.seed

If random.seed is NULL no random seed will be set. If random.seed is a number, it will be used as the random seed. This parameters should be used to get the same result for different runs.

Details

Tight clustering method is a resampling-evaluated clustering method that aims to directly identify tight clusters in a high-dimensional complex data set and allow a set of scattered objects without being clustered. The method was originally developed for gene cluster analysis in microarray data but can be applied in any complex data. The most important parameter is k.min. A large k.min results in smaller and tighter clusters. Normally k.min>=target+5 is suggested. All other parameters do not affect the quality of final clustering results too much and are suggested to remain unchanged.

Value

Returned value is a "tight.clust" object (list). The first element is the original data matrix. The second element is a vector representing the cluster identity (-1: scattered gene set; 1: the first cluster; 2: the second cluster; ...). The third element is a vector of the size of each tight cluster.

Author(s)

Chi Song <[e-mail]>

References

George C. Tseng and Wing H. Wong. (2005) Tight Clustering: A Resampling-based Approach for Identifying Stable and Tight Patterns in Data. Biometrics.61:10-16.

Examples

## load the test dataset
data(tclust.test.data)
## find 10 tight clusters
ptm<-proc.time()
## k.min=25, tighter clusters will be found
## target=1 is used to save time, target=10 is recommended
tclust1<-tight.clust(tclust.test.data$Data, target=1, k.min=25, random.seed=12345)
proc.time()-ptm
## plot the heat map of cluster result
plot(tclust1)
## write the cluster result
write.tight.clust(tclust1)
ptm<-proc.time()
## k.min=10, looser clusters will be found
## target=1 is used to save time, target=5 is recommended
tclust2<-tight.clust(tclust.test.data$Data, target=1, k.min=10, random.seed=12345)
proc.time()-ptm
## plot the heat map of cluster result
plot(tclust2)
## write the cluster result
write.tight.clust(tclust2)