Algorithms for bounded-error correlation of high dimensional data in microarray experiments

The problem of clustering continuous valued data has been well studied in literature. Its application to microarray analysis relies on such algorithms as k-means, dimensionality reduction techniques, and graph-based approaches for building dendrograms of sample data. In contrast, similar problems for discrete-attributed data are relatively unexplored. An instance of analysis of discrete-attributed data arises in detecting co-regulated samples in microarrays. In this paper, we present an algorithm and a software framework, PROXIMUS, for error-bounded clustering of high-dimensional discrete attributed datasets in the context of extracting co-regulated samples from microarray data. We show that PROXIMUS delivers outstanding performance in extracting accurate patterns of gene-expression.