Bayesian Methods for Complex Biological Data: Spatial Models, Regression, and Networks
Conference
Abstract
Modern biological studies generate complex data in which relationships vary across tissue locations, molecular features, and patient groups. This session presents three Bayesian approaches to learning from such structure. The first develops probabilistic models for spatial imaging in cancer, with applications to genomic networks, cellular interactions in the tumor microenvironment, and patient outcomes. The second introduces a smoothed functional horseshoe prior for semiparametric regression, enabling selection among linear and nonlinear effects and distinct components of interactions while controlling smoothness. The third develops profile graphical models that identify conditional dependence relationships shared across groups and those specific to a patient profile. Together, the talks highlight advances in structured modeling, variable and network selection, and uncertainty quantification for complex biological data.