Characterizing the evolutionary architecture of complex disease within and across diverse populations

The past decade of genome-wide association studies (GWASs) has seen thousands of complex traits and diseases studied and identified thousands of reproducibly associated genetic variants. GWAS has helped characterize the complexity of common genetic architectures and shed light on the role of genetics in disease risk. A large body of works have demonstrated that risks of complex traits are highly enriched in functional regions of the genome, which indicates that risk is mediated through perturbed regulatory action on relevant susceptibility genes. Similarly, multiple recent works have found that disease risks are shaped by forces of natural selection, which kept the frequencies of deleterious alleles low in the population. Together, the functional mechanisms and their interplay with natural selection can be coupled under a general mechanism we refer to as the evolutionary architecture. Current frameworks to infer the evolutionary architecture for common complex diseases are only applicable to relatively homogenous populations. Several recent works have demonstrated that integrating multi-population GWAS data substantially improves statistical power to identify causal factors underlying complex traits and diseases due to the increased heterogeneity in allele frequencies. Current approaches evolutionary architecture are unable to appropriately model the heterogeneity across populations with respect to allele frequencies and linkage disequilibrium. Similarly, the resolution of these methods is currently limited to complex diseases and phenotypes, whose inferred architectures, while informative, fail to describe regulatory network mechanisms that mediate risk. Methods capable of analyzing many molecular phenotypes simultaneously have the potential to identify shared architectures, and pinpoint core genes relevant for disease risk. Lastly, several works have shown that integrating functional information with GWAS substantially improves polygenic risk prediction. Together, these issues and opportunities highlight the need for new computational approaches that can scale to multiple populations and large-scale molecular phenotype catalogues while accounting for underlying heterogeneity and shared signals. Here, we propose novel approaches to integrate GWAS data from multiple populations and phenotypes to characterize the population-specific and shared evolutionary architectures. Importantly, our approaches run directly on summary data, which enables immediate large-scale analysis. We propose to apply our novel approaches to large-scale multi-population GWAS data. Together, our work will systematically characterize evolutionary architectures for complex diseases and molecular phenotypes and populations in a robust, open, and reproducible approach.