Fine-mapping complex traits in large-scale biobanks across diverse populations

Abstract

Identifying causal variants for complex traits is a major goal of human genetics research. Despite the great success of genome-wide association studies (GWAS) in locus discovery, individual causal variants in associated loci remain largely unresolved, limiting the biological inference possible from follow-up experimentation. In this dissertation, I present our fine-mapping analyses of complex traits in large-scale biobanks across diverse populations to create an atlas of causal variants.

We first fine-mapped complex traits using 361,194 European individuals from UK Biobank (UKBB) and gene expression using 49 tissues from GTEx (Chapter 1). We then extended our fine-mapping of complex traits to multiple populations, using 178,726 Japanese individuals from BioBank Japan and 271,341 Finnish individuals from FinnGen (Chapter 2). In total, we identified 4,518 variant-trait pairs with high posterior probability (> 0.9) of causality across the three populations. Aggregating data across populations enabled replication of 285 high-confidence variant-trait pairs as well as identification of 1,492 unique fine-mapped coding variants and 176 genes in which multiple independent coding variants influence the same trait. These results demonstrate that fine-mapping in diverse populations enables novel insights into the biology of complex traits by pinpointing high-confidence causal variants for further characterization.

Next, we investigated fine-mapping accuracy in GWAS meta-analysis (Chapter 3). We demonstrated that meta-analysis fine-mapping is substantially miscalibrated in simulations and proposed a novel quality-control method, SLALOM, that identifies suspicious loci for meta-analysis fine-mapping. Having validated SLALOM performance in simulations, we found widespread suspicious patterns in existing GWAS significant loci that call into question fine-mapping accuracy. We thus urge extreme caution when interpreting fine-mapping results from meta-analysis.

Finally, we introduce a new polygenic risk score (PRS) method, PolyPred, that improves cross-population polygenic prediction by combining a new fine-mapping-based predictor and a published BOLT-LMM predictor (Chapter 4). Leveraging estimated causal effects from fine-mapping enabled higher PRS transferability in non-European populations, achieving up to +32% improvement in prediction accuracy vs. BOLT-LMM using UKBB Africans.

Altogether, this work demonstrates key advances in fine-mapping complex traits across diverse populations and provides insights into further variant characterization as well as improved polygenic prediction based on fine-mapping.