A1531
Title: Uncovering high-dimensional genomic signals modulating biological networks via Gaussian graphical models
Authors: Samuel Anyaso-Samuel - National Cancer Institute (United States) [presenting]
Abstract: Gaussian graphical models (GGMs) are powerful tools for characterizing direct relationships among biological traits (e.g., gene expression, protein, microbial taxa) through partial correlation coefficients (PCCs). While traditional applications of GGMs focus on static biological networks, the influence of genomic factors (e.g., SNPs, mutations) on these networks remains underexplored. We propose a two-stage penalized regression framework to identify genomic regulators that modify trait-trait (edges) relationships. Our strategy first constructs a baseline network and then tests for genomic modifiers only among selected trait pairs, reducing computational and multiple testing burdens. Moreover, we develop an analytic procedure to control the false discovery rate (FDR) in regularized regressions, making large-scale analysis feasible. Through simulations, we demonstrate its computational efficiency, accurate Type-I error control, and high sensitivity. We demonstrate the utility of our framework in three case studies: identifying host genetic variants associated with oral microbiome networks, linking gut microbiome taxa to metabolite networks, and uncovering somatic mutations influencing gene expression networks in lung adenocarcinoma.
