2026 Sequencing Strategies for Population and Cancer Epidemiology Studies (SeqSPACE) : Submission #3

Abstract Summary Polygenic risk scores are widely used in disease risk stratification, but their accuracy varies across different ancestries. Recent methods leverage multi-ancestry data to improve accuracy in under-represented populations but require labelling individuals by ancestry. This poses practical challenges, as clinical decisions are typically not based on ancestry, and many individuals may not fit into a pre-specified ancestry group. We propose SPLENDID, a penalized regression framework for large-scale individual-level data that models genetic ancestry as a continuum to produce a unified prediction model without any ancestry labels. In extensive simulations and analysis in the All of Us Research Program (N=224,364) and UK Biobank (N=340,140), we show that SPLENDID significantly improved prediction accuracy over existing methods, particularly in non-European and admixed ancestries. By modeling genetic interactions with continuous ancestry, we further identified ancestry-differential effects in lipid and blood cell phenotypes that may explain limited transferability of existing PRS methods across ancestry groups. Finally, using a logistic regression extension of SPLENDID improved prediction of breast and prostate cancer by 6% and 9%, respectively, compared to current state-of-the-art PRS. Altogether, SPLENDID stands as a valuable tool for robust risk prediction across diverse populations, reduced health disparities in genetic research, and fairer clinical implementation.