On the categorization of a beta-uniform distribution and its application to the gene expression problem

Based on a set of cutting points we propose to categorize a betauniform (BUM) distribution, and term the categorized distribution as the C-BUM distribution. We study the categorization effects by comparing the performance of the maximum likelihood estimates for both models, for different sets of cutting points. Using the missing information principle we compute the BUM and C-BUM theoretical Fisher information matrices to assess loss of information. Finally, we evaluate the method to a published cancer gene expression data study.