Differential Shrinkage as a Way of Integrating Prior Knowledge in a Bayesian Model to Improve the Analysis of Genetic Association Studies

We propose a method of integrating external biological information about SNPs in a Bayesian hierarchical shrinkage model that jointly estimates SNP effects with the aim of increasing the power to detect variants in genetic association studies. Our method induces shrinkage on the SNP effects that is inversely proportional to prior information: SNPs with more information are subject to little shrinkage and more likely to be detected, while SNPs without prior information are strongly shrunk towards zero (no effect). The performance of the method was tested in a simulation study with 1000 datasets, each with 500 subjects and ∼1200 SNPs, divided in 10 Linkage Disequilibrium (LD) blocks. One LD block was simulated to be truly associated with the outcome. The method was further tested on an empirical example using BMI as the outcome and data from the European Community Respiratory Health Survey: 1,829 subjects and 2,614 SNPs from 30 blocks, 6 of which known to be truly associated with BMI. Prior knowledge was retrieved using the bioinformatic tool Dintor and incorporated in the model. The Bayesian model with inclusion of prior information outperformed the classical analysis. In the simulation study, the mean ranking of the true LD block was 2.8 for the Bayesian model vs. 3.6 for the classical analysis. Similarly, the mean ranking of the six true blocks in the empirical example was 8.3 vs. 11.7 in the classical analysis. These results suggest that our method represents a more powerful approach to detect new variants in genetic association studies.