Addressing Bias in Genomics: Genome-Wide Association Studies and Polygenic Risk Scores in Latinx Cohorts

Abstract

Genome-wide association analyses (GWAS) are an important tool for uncovering gene-trait associations and, increasingly, for genetic risk prediction. Differences in allele frequencies and linkage disequilibrium patterns, though, can lead to divergent GWAS results and predictive performances. Despite this, GWAS cohorts have been overwhelmingly of European ancestry. This means that the genetic architecture of many traits is poorly understood in many other populations and genetic risk models underperform for non-Europeans, potentially perpetuating existing health disparities. My current research seeks to shrink these biased knowledge gaps by improving representation of Latinx populations in GWAS data and identifying best practices for polygenic risk prediction in Latinx cohorts. In chapter 2, I performed the first GWAS in a Latinx Parkinson’s disease (PD) cohort (the Latin American Research Consortium on the Genetics of Parkinson’s Disease, or LARGE-PD), reaffirming the important of the SNCA locus for PD etiology and finding a potentially novel locus near the NRROS gene. Then, in chapter 3, I demonstrated the challenges of performing polygenic risk prediction for PD using European ancestry GWAS data. I found that the distribution of the European-ancestry polygenic risk score (PRS) exhibited shifts according to ancestry, potentially limiting its clinical utility. I then further explored the rs356182 variant in the SNCA locus as it is a major component of the PD PRS. By performing a haplotype analysis, I found that population-specific haplotypes with little shared variation other than rs356182 were associated with PD, providing orthogonal evidence that the variant rs256182 is functional. Finally, in chapter 4, I extended my focus to six additional complex traits such as height and type 2 diabetes and found that including non-European GWAS data improves predictive performance in Latinx cohorts even if that GWAS data did not contain Latinx subjects. In general, increasing PRS model complexity results in improved predictive performance, though the genetic architecture of a trait can alter this pattern, as is the case for prostate cancer. Overall, this work sheds light onto the genetic architecture of PD in Latinx cohorts, provides guidance for performing PRS-based prediction in non-Europeans, and helps move the field towards a more equitable future