A Distributed Markov Random Field Algorithm for Genome-Wide Association Studies

• Activity Number: 23
• Type: Topic Contributed
• Date/Time: Sunday, July 29, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Section on Government Statistics
• Author(s): Joong-Ho Won*+ and Ilana Belitskaya-Levy
• Companies: VAPAHCS and VA Palo Alto Cooperative Studies Program Coordinating Center
• Keywords: GWAS ; Markov random fields ; hidden Markov fields ; distributed computing

Abstract

Commonly used methods for genome-wide association studies (GWAS) test one single nucleotide polymorphism (SNP) or haplotype at a time and apply Bonferroni correction for adjusting for multiple comparisons, ignoring linkage disequilibrium (LD) among the loci. As a result, these methods often sacrifices statistical power. An alternative to the common approaches is to model the LDs using a Markov random field and treat the measurements as a noise observation of the field. While this approach captures the LD information effectively, the computational burden has been a hurdle in popularizing it. In this work, we propose a distributed computational method to estimate the association with uncertainty measure. This method is efficient and scales well with the number of SNPs. We demonstrate the performance of the proposed method using data from a case-control GWAS.