194 – Contributed Oral Poster Presentations: Biometrics Section
A Bayesian Approach for Two-Phase Designs in Regional Sequencing
Zhijian Chen
Samuel Lunenfeld Research Institute
Radu Craiu
University of Toronto
Shelley Bull
University of Toronto
Following detection of signals by genome-wide association studies (GWAS), investigators may choose to sequence some or all of the members of the GWAS sample to narrow down a set of potentially causal variants. This is known as a two-phase fine-mapping design. Additional efficiencies may be achieved if phase 2 fine mapping is carried out in multiple stages, with each stage comprised of a mutually exclusive subset. We consider a Bayesian approach to two-phase sampling that allows intermediate sampling time points and adaptive strategy. At each sampling point, we assess each sequence variant within a region by a Bayes factor that compares different genetic models, e.g., additive, dominant and recessive, and a null model containing no genetic effect. For variants in which no genetic model outperforms the others, we apply Bayesian model averaging to account for genetic model uncertainty. We assess the efficiency of this two-phase design in the discovery of true causal variants using posterior probabilities of association, and within a single region investigate the ability to narrow down a credible set that contains a true causal variant.