Brooke L Fridley1, Gregory D Jenkins. 1. Division of Biomedical Statistics and Informatics, Department of Health Sciences Research, Mayo Clinic College of Medicine, Rochester, MN 55905, USA. fridley.brooke @ mayo.edu
Abstract
OBJECTIVE: Numerous methods have been proposed to model the association between multiple single nucleotide polymorphisms (SNPs) and a phenotype. Often these methods do not explicitly model the information regarding the linkage disequilibrium (LD) between SNPs. Furthermore, many methods shrink the SNP effects towards zero, rather than to an unknown latent gene-level effect. METHODS: We outline the use of bayesian hierarchical models for gene-level analysis that incorporates LD information. Four different bayesian models, with either the inclusion or exclusion of LD information and 'shrinkage' of SNP effects to a latent gene-level effect or zero, were applied to a pharmacogenomic study and simulation STUDY. RESULTS: We observed that the inclusion of LD information resulted in increased precision in the SNP parameter estimates. The simulation study also demonstrated that the bayesian models were able to determine the simulated 'causative' variant more often with less false-positive associations as compared to commonly used multi-SNP analysis methods. CONCLUSIONS: Incorporating LD information in the analysis of multiple SNPs results in more precise estimates of SNP effects. In addition, the bayesian models are able to isolate the simulated 'causative' variant more often than commonly used regression modeling methods. Copyright 2010 S. Karger AG, Basel.
OBJECTIVE: Numerous methods have been proposed to model the association between multiple single nucleotide polymorphisms (SNPs) and a phenotype. Often these methods do not explicitly model the information regarding the linkage disequilibrium (LD) between SNPs. Furthermore, many methods shrink the SNP effects towards zero, rather than to an unknown latent gene-level effect. METHODS: We outline the use of bayesian hierarchical models for gene-level analysis that incorporates LD information. Four different bayesian models, with either the inclusion or exclusion of LD information and 'shrinkage' of SNP effects to a latent gene-level effect or zero, were applied to a pharmacogenomic study and simulation STUDY. RESULTS: We observed that the inclusion of LD information resulted in increased precision in the SNP parameter estimates. The simulation study also demonstrated that the bayesian models were able to determine the simulated 'causative' variant more often with less false-positive associations as compared to commonly used multi-SNP analysis methods. CONCLUSIONS: Incorporating LD information in the analysis of multiple SNPs results in more precise estimates of SNP effects. In addition, the bayesian models are able to isolate the simulated 'causative' variant more often than commonly used regression modeling methods. Copyright 2010 S. Karger AG, Basel.
Authors: Christopher S Carlson; Michael A Eberle; Mark J Rieder; Qian Yi; Leonid Kruglyak; Deborah A Nickerson Journal: Am J Hum Genet Date: 2003-12-15 Impact factor: 11.025