Several hierarchical Bayesian methods have been developed that combine prior information with genomic association data to help prioritize variants in various contexts [29], [30]. The main contribution of our approach is that we explicitly account for LD between SNPs which we can learn from external reference panels such as the 1000 Genomes. Additionally, because we do not take a fully Bayesian approach [30] (i.e. integrate over the entire hyper-parameter space), we are able to devise computationally efficient algorithms that allow our method to search over the ever-increasing number of functional annotations (e.g. ENCODE) to identify the most informative subset while retaining the ability to model multiple causal variants.
