A0456
Title: Mendelian randomization with pleiotropy through partially functional linear regression
Authors: Chi-Shian Dai - National Cheng Kung University (Taiwan) [presenting]
Abstract: A novel approach to Mendelian randomization (MR) is presented by employing functional instrumental variables (FIVs) for exploring causal relationships. Traditional MR relies on single nucleotide polymorphisms (SNPs) as instrumental variables, but these often suffer from weak instruments and pleiotropy. Functional two-stage least squares (F2TLS) is introduced, where SNPs are modeled as a function of their genomic positions within a gene, forming an FIV. This approach accounts for direct effects (pleiotropy), improving causal inference. A smoothness assumption is proposed to generalize the InSIDE requirement, ensuring the identifiability of the causal effect. An E-value type statistic is also introduced to gauge the assumption's robustness. The methodology is validated through theoretical analysis and simulations. It is then applied to the Alzheimer's Disease Neuroimaging Initiative (ADNI) database to investigate the link between gene expression and AD-related metrics. Results reveal a causal effect of APOC1 gene expression on cerebrospinal fluid beta-amyloid (ABETA) and phosphorylated tau (PTAU). F2TLS refines MR investigations and deepens the understanding of genetic determinants of disease. This demonstrates the power of integrating functional data into causal inference, advancing the utility of MR. This functional MR framework broadens the scope of genetic epidemiology and paves the way for robust causal discoveries.
