A bayesian approach for detection of multilocus interaction in case control studies

• Activity Number: 295
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Saonli Basu*+
• Companies: University of Minnesota
• Keywords: gene gene interaction ; bayesian approach

Abstract

Studying one single nucleotide polymorphism (SNP) at a time may not be sufficient to understand complex diseases. A SNP alone may have little effect on risk of disease, but together may increase the risk substantially. The joint behavior of genetic variants is often referred to as epistasis or multilocus interaction. We have proposed a bayesian partitioning model to detect such interaction. Our model clusters SNPs according to the direction of association, and computes, via Markov chain Monte Carlo, the posterior probability that each SNP is associated with the disease. Since the number of parameters to model interaction grows exponentially with the number of SNPs, we propose a pair-wise scoring approach to approximate high order interactions. We illustrate and compare our model with existing approaches and demonstrate the superiority of our model in detecting multilocus interaction.