Date of Award

January 2025

Document Type

Thesis

Degree Name

Master of Public Health (MPH)

Department

School of Public Health

First Advisor

Hongyu Zhao

Abstract

Polygenic risk scores (PRS) hold immense promise for personalized medicine but suffer from reduced accuracy in non-European populations, hindering equitable application. We introduce SWIFT (Strong-signal Weighted PRS Integration for Fine-Tuning), a novel framework to enhance cross-population PRS prediction by optimally combining base scores from diverse ancestries using only GWAS summary statistics. SWIFT employs a GWAS subsampling technique based on data fission to generate internal training and tuning sets, circumventing the need for external validation cohorts. A key innovation is its focus on strong genetic signals (p < 0.1) for estimating robust combination weights via cross-validation, which mitigates overfitting and significantly reduces computational burden. We evaluated SWIFT across 22 complex traits using data from five populations (EUR, AFR, AMR, EAS, SAS) and independent validation cohorts (UK Biobank, All of Us). Applied to state-of-the art base methods (PRS-CSx, X-Wing, JointPRS), SWIFT achieved predictive accuracy comparable or superior to the recent LEOPARD integration method and baseline implementations. Crucially, SWIFT demonstrated a 3.5- to 9.5-fold reduction in computational runtime compared to LEOPARD and exhibited robust performance even when using a single, unified LD reference panel. SWIFT provides an efficient, robust, and practical approach to improve the accuracy and applicability of PRS across diverse populations, representing a significant step towards more equitable genomic medicine.

Comments

This thesis is restricted to Yale network users only. It will be made publicly available on 12/16/2025

Download

Share

COinS