A Bayesian model selection method for multiple mediator analysis in genetic association studies

In exploring the genetic basis of biological traits, we frequently need to study how genetic variations affect phenotypic characteristics through different molecular levels (such as RNA, proteins). When we discover multiple phenotypic traits associated with the same genetic locus, such genetic associations may reflect biological mechanisms: one phenotype serves as an intermediate step that transmits the genetic effect to another phenotype. To explore the nature of these biological mechanisms, we employ mediation analysis as a statistical inference tool to dissect these biological relationships. We assessed the application of Bayesian model selection method in multiple mediation analysis to identify and quantify complex multi-mediator effects. This methodology extends beyond the constraints of traditional single-mediator analysis, enabling simultaneous evaluation of how multiple mediators influence phenotypic traits. We implemented this methodology in an R package, bmediatR. We compared the Bayesian model selection approach with conventional methods (Sobel test), and demonstrated its superior performance in identifying complex mediation relationships through simulation studies.