Benchmarking High-Dimensional Mediation Methods in Epigenome-Wide Studies
Bioinformatics Advances · 30.10.2026
DOI: 10.1093/bioadv/vbag331 ↗Kurz gefasst
Fachartikel in der Zeitschrift Bioinformatics Advances, online erschienen am 30.10.2026. Verfasst von Wan-Yu Lin. Beteiligte Einrichtung: Institute of Health Data Analytics and Statistics, College of Public Health, National Taiwan University, Taipei, 100025, Taiwan. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Motivation High-dimensional mediation analysis is increasingly used in omics studies to investigate molecular mediators of exposure–outcome relationships. In epigenome-wide studies, the number of CpG sites often greatly exceeds the sample size, while strong correlations among sites challenge reliable mediator selection and effect estimation. We systematically benchmarked four high-dimensional mediation methods—HIMA, HIMA2, HDMA, and a strategy combining high-dimensional mediation testing with stepwise regression (HDMTS)—under simulation settings varying signal strength, correlation structure, and the mediator-to-sample ratio (p/N). Results HIMA and HDMA consistently provided stringent false discovery proportion (FDP) control, whereas HIMA2 generally controlled FDP, with some inflation observed when the p/N ratio was high and mediator errors were independent. HDMTS showed greater sensitivity but exhibited reduced FDP control, particularly when mediator–outcome associations were weak and residual correlations among mediators were strong. Application to epigenome-wide DNA methylation data further demonstrated differences in mediator selection across methods and identified biologically plausible candidate mediators. Overall, our results highlight important trade-offs among false discovery control, sensitivity, and estimation accuracy and provide practical guidance for selecting high-dimensional mediation methods in epigenome-wide studies. Availability and implementation The HDMTS code is available at https://github.com/WanYuLin/HDMTS, and the simulation code is available at https://github.com/WanYuLin/HDMTS/tree/main/Simulations.
Autor:innen und Institutionen
- Institute of Health Data Analytics and Statistics, College of Public Health, National Taiwan University, Taipei, 100025, Taiwan.
- Master of Public Health Program, College of Public Health, National Taiwan University, Taipei, 100025, Taiwan.
ORCID-Profile: Wan-Yu Lin
Zitieren
APA
Lin, W. Y. (2026). Benchmarking High-Dimensional Mediation Methods in Epigenome-Wide Studies. Bioinformatics Advances. https://doi.org/10.1093/bioadv/vbag331
Vancouver
Lin WY. Benchmarking High-Dimensional Mediation Methods in Epigenome-Wide Studies. Bioinformatics Advances. 2026. doi:10.1093/bioadv/vbag331
BibTeX
@article{Lin2026,
author = {Lin, Wan-Yu},
title = {Benchmarking High-Dimensional Mediation Methods in Epigenome-Wide Studies},
journal = {Bioinformatics Advances},
year = {2026},
doi = {10.1093/bioadv/vbag331}
}
EUVIMED