Abstract
While most signals identified by genome-wide association studies (GWAS) are shared across populations, the growing size and diversity of GWAS datasets provides evidence that a subset of signals are ancestry-specific. Yet, characterizing these signals remains challenging, since the underlying causal variants are often unknown. Statistical fine-mapping aims to identify candidate causal variants, but struggles to distinguish between variants in high linkage disequilibrium (LD). Multi-study fine-mapping methods can improve resolution by leveraging population-specific LD patterns, but typically assume causal variants are shared and/or polymorphic across studies, making it challenging to study ancestry-specific contributions. To overcome these limitations, we introduce PIPSORT, a multi-study fine-mapping method which simultaneously detects both shared and ancestry-specific signals and quantifies evidence of signal sharing across studies. We applied PIPSORT to fine-map platelet count and LDL cholesterol (LDL-C) in individuals of primarily African vs. European ancestry in the UK Biobank (UKB) and All of Us (AoU) datasets. Due to the bias of these datasets toward Europeans (94% in UKB, 49% in AoU), most trait-associated regions identified have strong signals in Europeans. We estimate 89%-99% of these regions are shared with Africans, but detect dozens of examples of ancestry-specific signals. Of these, 10, including known African-specific missense variants in MPL (platelet count) and PCSK9 (LDL-C), could only be confidently detected in the more diverse AoU cohort. We additionally applied PIPSORT to fine-map schizophrenia signals in East Asians vs. Europeans, which identified multiple ancestry-specific signals including a known missense variant in SLC39A8. Finally, we leveraged the high degree of admixture within AoU to identify specific signals for platelet count and LDL-C that are driven by interaction with local vs. global ancestry. Overall, our finding that multiple strong ancestry-specific signals could be identified for all traits studied provides novel insights into the genetic architecture of complex traits in different populations and has important implications for development of future multi-ancestry methods for complex trait analysis.