A0750
Title: Robust and flexible high-dimensional causal mediation model for DNA methylation studies
Authors: An-Shun Tai - National Cheng Kung University (Taiwan) [presenting]
Abstract: In the pathogenesis of diseases, DNA methylation (DNAm) markers play a pivotal role in influencing gene expression and engaging in diverse biological processes. Given the extensive number of DNAm markers, exceeding half a million, implementing a high-dimensional mediation model is necessary to identify the activated DNAm markers within the mediation pathway and assess their mediation effects. Most existing high-dimensional mediation models necessitate stringent assumptions, including correctly prespecifying the mediation relationship and determining all outcomes, mediators, and exposure models. However, fulfilling these assumptions is challenging in the context of high-dimensional mediators. A novel Bayesian estimation procedure is studied for interventional mediation effects, offering robustness against model misspecification and flexibility in prespecifying the mediation structure. Spike-and-slab priors are employed to integrate Bayesian variable selection into the modeling process. The proposed method is demonstrated using publicly available genome-wide array-based cancer studies to estimate the causal effects mediated through DNAm.
