How does gene expression clustering work?

PRIMERHow does gene expression clustering work?.com/naturebiotechnologyPatrik DhaeseleerClustering is often one of the first steps in gene expression analysis. How do clustering algorithms work, which ones

should we use and what can we expect from them?Our ability to gather genome-wide expression

data has far outstripped the ability of our puny

human brains to process the raw data. We can

distill the data down to a more comprehensible

level by subdividing the genes into a smaller

number of categories and then analyzing those.

This is where clustering comes in.The goal of clustering is to subdivide a set

of items (in our case, genes) in such a way that

similar items fall into the same cluster, whereas

dissimilar items fall in different clusters. This

brings up two questions: first, how do we

decide what is similar; and second, how do we

use this to cluster the items? The fact that these

two questions can often be answered independently contributes to the bewildering variety

of clustering algorithms.Gene expression clustering allows an openended exploration of the data, without getting lost among the thousands of individual

genes. Beyond simple visualization, there are

also some important computational applications for gene clusters. For example, Tavazoie

et al.1 used clustering to identify cis-regulatory

sequences in the promoters of tightly coexpressed genes. Gene expression clusters also

tend to be significantly enriched for specific

functional categorieswhich may be used to

infer a functional role for unknown genes in

the same cluster.In this primer, I focus specifically on clustering genes that show similar expression patterns across a number of samples, rather than

clustering the samples themselves (or both). I

hope to leave you with some understanding

of clustering in general and three of the more

popular algorithms in particular. Where possible, I also attempt to provide some practical

guidelines for applying cluster analysis to your

own gene expression data sets.A few important caveatsBefore we dig into some of the methods in

use for gene expression data, a few words of

caution to the reader, practitioner or aspiring

algorithm developer: It is easyand temptingto invent yet

another clustering algorithm. There are hundreds of published clustering algorithms,

dozens of which have been applied to gene http://www.nature2005 Nature Publishing Group aExperiment 2bExperiment 1Experiment 2Experiment 1cdBob CrimiPatrik Dhaeseleer is in the...