A1456
Title: Mendelian randomization with pleiotropy through partially functional linear regression
Authors: Cai Li - St. Jude Children's Research Hospital (United States) [presenting]
Abstract: A novel Mendelian randomization (MR) framework is proposed, that models instrumental variables as random functions to account for pleiotropy through a functional partially linear regression. Unlike conventional MR methods that treat SNPs individually, this approach leverages information across entire genes, thereby capturing correlations among SNPs and strengthening signals. The method incorporates a roughness penalty that both respects the structure of functional effects and serves as a regularization to facilitate identifiability of causal effects in the presence of pleiotropy. Building on this, a "smoothness" assumption is introduced, which generalizes the InSIDE (Instrument Strength Independent of Direct Effect) assumption, to further guarantee causal identifiability. For estimation, a penalized partially functional linear regression approach is proposed, implemented as a one-step generalized method of moments (GMM) procedure. This framework enables inference on both causal effects and functional direct effects. The utility of the method is illustrated in uncovering causal relationships between gene expressions and Alzheimer's disease biomarkers. Simulation studies demonstrate that our approach performs favorably compared with state-of-the-art MR methods.
