Bayesian Approaches to Prognostic Gene Signature Identification from Microarray Data in a Functional Domain: Applications in Ovarian Cancer Survival Data

• Activity Number: 455
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Miranda Lynch*+
• Companies: University of Connecticut Health Center
• Keywords: Gene expression ; Gene signature ; Ovarian cancer ; Survival data ; Bayesian methods

Abstract

Iron metabolism plays an important role in tumor growth in some types of cancer, and there has been interest in characterizing the molecular pathways by which changes in iron regulation affect malignancy. Comprehending relevant pathways and how they impact on prognosis requires identification of gene signatures within this functional category that can discriminate patient survival outcomes. A variety of methods have been proposed to identify prognostic gene signatures. Because we are interested in identifying a gene signature within a functionally related set of genes, many of which may be encoding proteins on the same or similar pathways, problems with correlation are especially acute and require special handling. This work presents a two-stage approach developed in a Bayesian framework that identifies an iron regulatory prognostic gene signature using clinical survival data, and addresses how it accommodates the correlation problem. We demonstrate the performance of the method, as well as apply the method to sets of publicly available ovarian cancer expression data. We further examine the impact of the identified genes across ovarian cancer molecular subtypes.