Development of Genomic Prediction Models for Personalized Medicine

Abstract

Populations of patients are heterogeneous due to differences in genetic pre-dispositions or disease characteristics. Personalized medicine aims at using molecular signatures of individual patients for better matching of disease with specific therapies for selection of the treatment that is most likely to be optimal for each patient. The success of personalized medicine depends on having accurate prediction models that can identify patients who can benefit from a targeted therapy prior to treatment. Development of a prediction model involves identification of sets of markers (signatures) based on patient’s genomic profiles and/or disease characteristics that are associated treatment outcomes. This talk presents an approach to identifying sets of markers sharing compatible patterns across a subset of patients via bicluster analysis. Classification is then developed to discriminate between different biologic phenotypes for distinguishing subpopulations. The approach is illustrated by applications to one synthetic data and real datasets.