A method to improve capture of causal genetics and by extension, cross-population portability when constructing polygenic scores

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ABSTRACT Polygenic scores (PGS) can predict disease risk in a population or an individual using genetic data and are poised to improve clinical care by making personalized preventive medicine a reality. Unfortunately, current methods of PGS are less accurate predictors across populations of non-European ancestry as a consequence of Eurocentric biases in genome-wide association studies (GWAS). PGS prediction accuracy can also vary within just a single population due to differences in the environment experienced by individuals. These biases in prediction both within and between populations severely limit the applicability of PGS. Unlike other forms of medical inequity which benefit one population while harming or neglecting another (e.g., prescription drugs such as some long-acting β2-agonist asthma treatments which exacerbate illness in African-ancestry populations), PGS perform ubiquitously better across European populations, leading clinical applications to systematically benefit individuals of European descent and neglecting the rest of the world. Despite this systematic bias, clinical trials of PGS are underway with applications in breast and prostate cancers, type I diabetes, and cardiovascular disease. Interest in PGS is only growing despite its limitations, so I propose to develop methods that can aid in mitigating some of the harm caused by premature applications of PGS. In Aim 1 I will build and apply PGS(C) a method of PGS which can account for the effects of a binarized context (e.g., sex) on a trait by incorporating gene by context interactions (GxC) into my PGS model. I will apply this model to improving prediction of sexually dimorphic traits such as major depression and Alzheimer’s disease witin multiple diverse datasets. This will yield a more portable PGS better able to predict disease risk in varied populations, incorporating biological variability, and gene by environment (GxE) interactions into prediction. In Aim 2 I will extend this method to incorporate continuous contexts (e.g., ancestry, environment, age, etc.) into prediction. Additionally, I will compare my novel PGS(C) method to existing state-of-the-art PGS methods to itdentify when each method mostly accurately predicts a trait while minimizing loss in portability. This work is a concerted effort to improve PGS portability, a crucial step in constructing a score that can bridge existing gaps in genetic medicine negatively impacting diverse and underrepresented study populations.