RiVIERA-MT: A Bayesian model to infer risk variants in related traits using summary statistics and functional genomic annotations

Fine-mapping causal variants is challenging due to linkage disequilibrium and the lack 11 of interpretation of noncoding mutations. Existing fine-mapping methods do not scale 12 well on inferring multiple causal variants per locus and causal variants across multiple 13 related diseases. Moreover, many complex traits are not only genetically related but 14 also potentially share causal mechanisms. We develop a novel integrative Bayesian 15 fine-mapping model named RiVIERA-MT. The key features of RiVIERA-MT include 16 1) ability to model epigenomic covariance of multiple related traits; 2) efficient pos17 terior inference of causal configuration; 3) efficient full Bayesian inference of enrich18 ment parameters, allowing incorporation of large number of functional annotations; 19 4) simultaneously modeling the underlying heritability parameters. We conducted a 20 comprehensive simulation studies using 1000 Genome and ENCODE/Roadmap epige21 nomic data to demonstrate that RiVIERA-MT compares quite favorably with existing 22 methods. In particular, the efficient inference of multiple causal variants per locus 23 led to significantly improved estimation of causal posterior and functional enrichments 24 compared to the state-of-the-art fine-mapping methods. Furthermore, joint modeling 25 multiple traits confers further improvement over the single-trait mode of the same 26 model, which is attributable to the more robust estimation of the enrichment parame27 ters especially when the annotation measurements (i.e., ChIP-seq) themselves are noisy. 28 We applied RiVIERA-MT to separately and jointly model 7 well-powered GWAS traits 29 including body mass index, coronary artery disease, four lipid traits, and type 2 di30 abetes. To leverage potential tissue-specific epigenomic co-enrichments among these 31 traits, we harness 52 baseline functional annotations and 220 tissue-specific epigenomic 32 annotations from well-characterized cell types compiled from ENCODE/Roadmap con33 sortium. Overall, we observed an improved enrichments for GTEx whole blood and 34