|
Abstract:
|
In identifying subgroups of a heterogeneous disease or condition, it is often important to identify both the observations and the characteristics which differ between subgroups. For instance, it may be that there is a subgroup of individuals with a certain disease who differ from the rest of the population based on the expression profile for a specific set of genes. We can represent the subgroup of individuals and genes as a bicluster, a submatrix such that the features and observations in the bicluster differ from those in the remaining data matrix. We have developed a novel two-step method, SC-Biclust, to identify biclusters. In the first step observations in the bicluster are identified to maximize the sum of the weighted between cluster feature differences. In the second step, features in the bicluster are identified based on their contribution to the clustering of the observations. SC-Biclust can be used to identify biclusters which differ based on feature means, feature variances, or more general differences. We illustrate the utility of SC-Biclust through simulation studies and real data applications.
|
