Improving Multi‐Trait Polygenic Risk Score Prediction of Drug Response Using Fine‐Mapping and Ensemble Learning
Statistics in Medicine · 2026-11 · 45 (25-27)
DOI: 10.1002/sim.70826 ↗Kurz gefasst
Fachartikel in der Zeitschrift Statistics in Medicine, online erschienen am 11.11.2026. Verfasst von Song Zhai und 2 weiteren Autor:innen. Beteiligte Einrichtung: Biostatistics and Research Decision Sciences, Merck & Co., Inc., Rahway, New Jersey, USA. Hauptthemen laut MEDLINE: Multifactorial Inheritance. Die Arbeit nennt 42 Literaturangaben.
Abstract (Originalsprache)
Multi‐trait Polygenic Risk Score (mtPRS) methods improve prediction accuracy compared to single‐trait PRS (stPRS) by integrating genetically correlated traits. Current mtPRS methods primarily use three approaches: (1) aggregating genome‐wide association studies (GWAS) summary statistics; (2) combining individual PRSs with weights from Principal Component Analysis (PCA); or (3) combining individual PRSs with multivariate or penalized regression‐derived weights. However, existing mtPRS methods face two main challenges: (1) selected variants are often non‐causal, lowering prediction power; and (2) no method is universally superior across diverse genetic architectures and multi‐trait correlation structures. To improve prediction performance and robustness, we propose mtPRS‐FIMEL (multi‐trait Polygenic Risk Score using FIne‐Mapping and Ensemble Learning), an ensemble learning approach that integrates fine‐mapping with multi‐trait PRS construction. mtPRS‐FIMEL computes mtPRS by combining multiple complementary mtPRS and stPRS methods and strategically leveraging causal variants identified through fine‐mapping analysis. Specifically, it uses GWAS summary statistics to estimate causal effects and learn optimal weights for a set of PRSs from complementary mtPRS (including mtPRS‐PCA, a PCA based method, and mtPRS‐ML, an elastic‐net based machine learning method, and stPRS). Our extensive simulations demonstrate that mtPRS‐FIMEL outperforms existing mtPRS and stPRS methods in terms of prediction accuracy, PRS association strength, and patient stratification performance in the simulation settings examined. Validation with PGx GWAS data from a randomized clinical trial successfully predicted the drug response of low‐density lipoprotein cholesterol (LDL‐C) change from baseline, while integrating three LDL‐C‐related traits (LDL, HDL, and Triglycerides) from disease GWAS, confirming its promising performance and robustness via a proof‐of‐concept internally cross‐validated approach.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Substanzen
Autor:innen und Institutionen
- Biostatistics and Research Decision Sciences, Merck & Co., Inc., Rahway, New Jersey, USA.
- Data, AI and Genome Sciences, Merck & Co., Inc., Cambridge, Massachusetts, USA.
ORCID-Profile: Song Zhai, Bin Guo, Judong Shen
Zitieren
APA
Zhai, S., Guo, B., & Shen, J. (2026). Improving Multi‐Trait Polygenic Risk Score Prediction of Drug Response Using Fine‐Mapping and Ensemble Learning. Statistics in Medicine, 45(25-27). https://doi.org/10.1002/sim.70826
Vancouver
Zhai S, Guo B, Shen J. Improving Multi‐Trait Polygenic Risk Score Prediction of Drug Response Using Fine‐Mapping and Ensemble Learning. Statistics in Medicine. 2026;45(25-27). doi:10.1002/sim.70826
BibTeX
@article{Zhai2026,
author = {Zhai, Song and Guo, Bin and Shen, Judong},
title = {Improving Multi‐Trait Polygenic Risk Score Prediction of Drug Response Using Fine‐Mapping and Ensemble Learning},
journal = {Statistics in Medicine},
year = {2026},
volume = {45},
number = {25-27},
doi = {10.1002/sim.70826}
}
EUVIMED