maanova (version 1.42.0)

macluster: Clustering analysis for Microarray experiment

Description

This function bootstraps K-means or hierarchical clusters and builds a consensus tree (consensus group for K-means) from the bootstrap result.

Usage

macluster(anovaobj, term, idx.gene, what = c("gene", "sample"),  method = c("hc", "kmean"), dist.method = "correlation", hc.method = "ward", kmean.ngroups, n.perm = 100)

Arguments

anovaobj

The result object for fitting ANOVA model.

term

The factor (in formula) used in clustering. The expression level for this term will be used in clustering. This term has to correspond to the gene list, e.g, idx.gene in this function. The gene list should be the significant hits in testing this term.

idx.gene

A vector indicating the list of differentially expressed genes. The expression level of these genes will be used to construct the cluster.

what

What to be clustered, either gene or sample.

method

The clustering method. Right now hierarchical clustering ("hc") and K-means ("kmean") are available.

dist.method

Distance measure to be used in hierarchical clustering. Besides the methods listed in dist, there is a new method "correlation" (default). The "correlation" distance equals to (1 - $r^2$), where r is the sample correlation between observations.

hc.method

The agglomeration method to be used in hierarchical clustering. See hclust for detail.

kmean.ngroups

The number of groups for K-means cluster.

n.perm

Number of bootstraps. If it is 1, this function will cluster the observed data. If it is bigger than 1, a bootstrap will be performed.

Value

An object of class macluster.

Details

Normally after the F test, user can select a list of differentially expressed genes. The next step is to investigate the relationship among these genes. Using the expression levels of these genes, the user can cluster the genes or the samples using either hierarchical or K-means clustering algorithm. In order to evaluate the stability of the relationship, this function bootstraps the data, re-fits the model and recluster the genes/samples. Then for a certain number of bootstrap iterations, say, 1000, we have 1000 cluster results. We can use consensus to build the consensus tree from these 1000 trees.

Note that if you have a large number (say, more than 100) of genes/samples to cluster, hierarchical clustering could be very unstable. A slight change in the data can result in a big change in the tree structure. In that case, K-means will give better results.

See Also

hclust, kmeans, consensus

Examples

# load in data
data(abf1)
# fit the anova model
## Not run: 
# fit.fix = fitmaanova(abf1,formula = ~Strain)
# # test Strain effect 
# test.fix = matest(abf1, fit.fix, term="Strain",n.perm= 1000)
# # pick significant genes - pick the genes selected by Fs test
# idx <- volcano(test.fix)$idx.Fs
# # do k-means cluster on genes
# gene.cluster <- macluster(fit.fix, term="Strain", idx, what="gene", 
#    method="kmean", kmean.ngroups=5, n.perm=100)
# # get the consensus group
# consensus(gene.cluster, 0.5)
# 
# # HC cluster on samples
# sample.cluster <- macluster(fit.fix, term="Strain", idx, what="sample",method="hc")
# # get the consensus group
# consensus(sample.cluster, 0.5)## End(Not run)