Binarization of consensus partition matrices

Mainly in the context of gene clustering, the binarization of consensus partition matrices (Bi-CoPaM) was proposed by Abu-Jamous et al.^[1] as a method for consensus clustering. In contrast to other conventional clustering and ensemble clustering methods, Bi-CoPaM has the ability to combine the results of clustering the same set of genes from various microarray datasets and by using many clustering methods to produce one consensus result. Moreover, Bi-CoPaM relaxes conventional clustering constraints by allowing each gene to have any of the three possible eventualities – to be exclusively assigned to one and only one cluster (as any conventional clustering method does), to be simultaneously assigned to multiple clusters, or to be unassigned from all of the clusters. At the clusters level, clusters can be complementary (as in the case of conventional clustering), can be wide and overlapping, and can be tight and distinct while leaving many genes unassigned from all of them. The Bi-CoPaM method has not been designed to only allow for these three forms of clusters; it has also been provided with tuning parameters which can be used to tune the level of tightness and wideness of the clusters based on research requirements.

Complete description of the method is given in the publication in which it was proposed (Abu-Jamous et al 2013).^[1]

Applications

As the Bi-CoPaM specially meets many requirements of gene discovery studies, its current main applications are within this field of bioinformatics;^[2] though, it was defined in a completely independent manner such that it is applicable for any other clustering problem. For example, a recent experiment in which the Bi-CoPaM was applied over multiple yeast cell-cycle datasets revealed important information about a poorly characterised gene, CMR1/YDL156W, and about its relation with many other genes.^[3]

References