From Genomics Data to Causality: An Integrated Pipeline for Mendelian Randomization

Abstract

ABSTRACTBackgroundMendelian randomization (MR) has emerged as a valuable tool for causal inference in genetic epidemiology. Existing MR methods have issues related to pleiotropy and offer limited comprehensiveness. Here, we introduce an integrated MR analysis pipeline designed for GWAS summary statistics data. Our pipeline integrates feature selection, harmonization, and checkpoint mechanisms to improve the accuracy and reliability of MR analysis.MethodsIn classical GWAS, the p-value threshold usually does not guarantee to identify causal single-nucleotide polymorphisms (SNPs). In such cases, t-statistics can be considered as imperative and robust criteria for identifying causal SNPs. Therefore, in this study, we computed the t-statistic for all independent SNPs remained after linkage disequilibrium pruning. Next, prior to harmonization, we removed SNPs having a t-statistic below the average t-statistic value. Furthermore, our pipeline incorporates sensitivity analysis tests at each step to reduce the chances of directional pleiotropy.Result and ConclusionWe applied our pipeline to single-sample and two-sample MR study designs, encompassing diverse populations and a wide range of diseases. Our results demonstrate superior performance compared to existing MR methods. In conclusion, our research presents an integrated MR analysis pipeline that significantly enhances the accuracy and reliability of MR studies. By outperforming existing methods and providing comprehensive validation, this pipeline represents a valuable tool for researchers in genetics and epidemiology.