Identifying Rare Haplotypes Associated with Common Diseases Through Bayesian Regularization

• Activity Number: 591
• Type: Invited
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Shili Lin*+ and Swati Biswas
• Companies: The Ohio State University and University of North Texas
• Keywords: Rare variant ; common disease ; haplotype ; generalized linear model ; Bayesian regularization ; age-related macular degeneration

Abstract

Identifying rare variants associated with a common disease is a challenging problem. To detect associated haplotypes under case-control sampling design, we propose a Bayesian regularized approach based on a generalized linear model (rGLM-Bayes). rGLM-Bayes employs a retrospective likelihood, an appropriate formulation for a case-control design. The regularization in rGLM-Bayes is carried out through a logistic version of Bayesian Lasso, in which the coefficients are penalized using appropriate prior distributions. The penalization of coefficients results in weeding out haplotypes that are not associated with the disease so that the associated ones, especially those that are rare, can stand out and be accounted for more precisely. We have conducted simulations under various settings involving different combinations of truly associated haplotypes to investigate the rGLM-Bayes approach and compare with an unregularized method. Our results show that rGLM-Bayes is much more powerful in identifying rare associated haplotypes when the false positive rates for both approaches are kept the same. Application to data for age-related macular degeneration identified several rare haplotypes.