A1504
Title: Sample-specific Multiomic Association networks Using Gaussian graphical models
Authors: Enakshi Saha - University of South Carolina (United States) [presenting]
Abstract: Gaussian Graphical Models (GGM) provide an invaluable tool for studying the interaction network between multiple omics modalities. However, existing methods estimate a single network that approximates the average conditional dependence structure across the entire population and fail to recognize individual-specific variability. To overcome this limitation, we propose an empirical Bayesian model, SMAUG (Sample-specific Multiomic Association networks Using Gaussian graphical models), that recognizes individual-specific heterogeneity in molecular dependence by estimating sample-specific GGMs. By employing data-driven Individual-specific conjugate priors, SMAUG provides a scalable tool for deciphering variability in disease mechanisms across sex, age and other clinical variables, thereby providing a more nuanced understanding of diseases. In addition, SMAUG, being a partial-correlation-based method, is better suited to distinguish between direct molecular dependence and spurious correlations, compared to existing methods for sample-specific network inference that employ Pearson's correlation as their foundation. We demonstrate the efficacy of SMAUG using simulated and real datasets.
