Clustering Approaches for Gene Expression Data

Since the early days of the development of microarray technologies, a
wide range of existing clustering algorithms have been used, and novel new
approaches have been developed for clustering gene expression data sets.
The most effective traditional clustering algorithms are based either on the
group-average variation of the agglomerative clustering methodology, or on the
K-means approach applied to unit-length gene or condition expression vectors.
Unlike other applications of clustering in life sciences, such as the construction
of phylogenetic trees, or guide trees for multiple sequence alignment, there is
no biological reason that justifi es that the structure of the correct clustering
solution is in the form of a tree. Thus, agglomerative solutions are inherently
suboptimal when compared to partitional approaches, which allow for a wider
range of feasible solutions at various levels of cluster granularity. However,
the agglomerative solutions do tend to produce reasonable and biologically
meaningful results and allow easy visualization of the relationships between
the various genes and/or conditions in the experiments.