Efficient Mendelian randomization analysis with self-adaptive determination of sample structure and multiple pleiotropic effects.

Abstract

Mendelian randomization (MR) has emerged as a highly valuable tool for inferring the causal effects of exposures on outcomes in observational studies using genetic variants, typically single-nucleotide polymorphisms (SNPs), as instrumental variables (IVs). Standard MR typically involves three steps: inputs of genome-wide association studies (GWASs) for both exposure and outcome, determination of IVs, and inference of causal effects. However, existing methods fail to simultaneously account for characteristics of GWAS data, uncertainty surrounding the validity of SNPs as IVs, and efficiency of estimating and testing the causal effect. Here, we developed MR method with self-adaptive determination of sample structure and multiple pleiotropic effects (MAPLE), a method for effective MR analysis. MAPLE utilizes correlated SNPs, self-adaptively accounts for the sample structure and the uncertainty that these correlated SNPs may exhibit multiple pleiotropic effects, and relies on a maximum-likelihood framework to infer the causal effects and obtain calibrated p values. We illustrate the advantage of MAPLE through comprehensively realistic simulations, where MAPLE, compared with another eight MR methods, shows calibrated type I error control and reduces false positives while being more powerful. In three types of lipid-trait-centric MR analyses in UK Biobank, MAPLE produces the most accurate causal-effect estimates in positive-control analyses evaluating the causal effect of each lipid trait on itself; reduces the false positives by 12.5% on average compared with existing methods in negative-control analyses investigating the causal effects of lipid traits on hair color and skin color; and highlights the causal effects of physical activity, alcohol, and smoking on lipid profiles in factor-screening analyses involving 412 trait pairs.