Data Complexity in Clustering Analysis of Gene Microarray Expression Profiles

The increasing application of microarray technology is generating large amounts of high dimensional gene expression data. Genes participating in the same biological process tend to have similar expression patterns, and clustering is one of the most useful and efficient methods for identifying these patterns. Due to the complexity of microarray profiles, there are some limitations in directly applying traditional clustering techniques to the microarray data. Recently, researchers have proposed clustering algorithms custom tailored to overcome their limitations for microarray analysis. In this chapter, we first introduce the microarray technique. Next, we review seven representative clustering algorithms: K-means, quality-based clustering, hierarchical agglomerative clustering, self-organizing neural network-based clustering, graph-theory-based clustering, model-based clustering, and subspace clustering. All these algorithms have shown their applicability to the microarray profiles. We also survey several criteria for evaluating clustering results. Biology plays an important role in the evaluation of clustering results. We discuss possible research directions to equip clustering techniques with underlying biological interpretations for better microarray profile analysis.