Bayesian fine mapping of complex disease genes utilising estimates of the number of yet-to-be-discovered disease-specific SNPs

Fine mapping studies aim to prioritize causal variants in complex diseases within genome-wide association studies (GWAS) regions. Bayesian approaches have been widely used in recent fine mapping studies because of the advantage they have in overcoming the limitations of the frequentist approach. The commonly used prior distribution on the causal single-nucleotide polymorphism (SNP) effect size is a Normal distribution with mean zero. Previous studies have shown that the posterior distribution and Bayes factor are both highly sensitive to the Normal prior variance, it is therefore reasonable to assume that posterior summaries are also sensitive to the parametric form of the prior. We show that the Laplace prior for the SNP effect size better reflects both the effect sizes observed in breast cancer GWAS top hits, and the number of yet-to-be-discovered SNPs, than the Normal prior. We estimate the prior parameters from the GWAS top hits and develop single-SNP and multi-SNP approaches for the Laplace prior. We compare our approaches with other existing fine mapping methods using simulated data from HAPGEN and real data from iCOGs. Our analysis shows that the Laplace prior performs better than the current gold standard multi-SNP fine mapping method in terms of causal SNP ranks.