|
Abstract:
|
For heterogeneous diseases, one of the most commonly used approaches to providing accurate and informative prognoses is to define subtypes. Subtypes traditionally have been defined using clinical parameters. The profusion of microarray data now available, however, makes it possible to define subtypes using molecular profiles. Based upon the idea of reproducibility, our method for validating the subtypes of a disease uses two independent microarray datasets and a new quantity called the in-group proportion (see the R package clusterRepro). The method is essentially a permutation test and several procedures for generating the null distribution of in-group proportions are discussed. When the method was applied to breast cancer tumor tissue microarray datasets using centroids previously defined, it provided evidence for the reproducibility of the basal, ERBB2+, and luminal B subtypes.
|
