Mendelian Randomization Methods Aiming At Correcting Horizontal Pleiotropy And Its Applications In Causal Inference

Objectives:In Mendelian Randomization studies,importing more genetic variables as instrumental variables can improve the effectiveness of causal estimation.However,more genetic variables may bring in more horizontal pleiotropy which can induce bias of the causality in the meantime.In this study,we simulated four Mendelian randomization methods(IVW method,MR-Egger regression method,WME method,and MR-PRESSO method)in order to compare their performance of causal inference when there were different types or proportions of horizontal pleiotropy.Finally,we applied the above methods to explore the causal relationship between sleep-related traits and Type 2 Diabetes Mellitus.Methods:Base on the principles of IVW,WME,and MR-Egger,and MR-PRESSO,we set up five simulation situations to exploring the influence of type and proportion of horizontal pleiotropy in each method respectively.Scenario one was a zero hypothesis without horizontal pleiotropy.In scenario two and scenario three,uncorrelated pleiotropy was set under different proportions in order to evaluate the performance of each MR method.In scenario four,different ratios of pleiotropy were set to evaluating the performance of each method when only corrected horizontal pleiotropy existed.Scenario five setup co-exist correlated pleiotropy and uncorrelated pleiotropy in equal proportions,aiming to evaluate whether four methods were stable in situation with two types of horizontal pleiotropy.And we used mean causal effect value,standard error,mean square error,Type ? error rate,and power as indicators to evaluate the advantages and disadvantages of four MR methods under different parameter settings.In application study,we used MR-PRESSO method as primary analysis while the other three methods were used as sensitivity analyses to explore the causal association between sleep related traits and T2 DM.Results:According to simulation results,we found when all the instrument variables were valid,in addition to MR-Egger regression method always with a high Type 1 error rates,each method is stable and can give an unbiased and consistent estimate.When there is balanced uncorrected pleiotropy,results from the MR-PRESSO and IVW method are stable with lower Type 1 error rates and less biased estimates.When there is directional uncorrected pleiotropy,the estimates from MR-Egger are less biased with the increasing proportion of invalid genetic variants,estimates from MR-PRESSO regression are close to unbiased.However,the Type 1 error rates are inflated.Corrected pleiotropy has a big influence on IVW and MR-Egger regression methods.Estimates from MR-Egger regression are close to unbiased in corrected pleiotropy situation.According to the results of the application study,we found that there was a causal relationship between insomnia and T2DM(OR:1.14,95%CI:1.08-1.20,P=2.11E-06),but we found no evidence of a causal relationship between the other sleep related traits and T2DM(P>0.01).Conclusions:Comparing with the other three MR methods,estimates from MR-PRESSO are close to unbiased in most simulation cases,which indicated that the MR-PRESSO method should be more widely used in the Two-Samples MR analysis to infer the causal relationship.In the application study,we utilized the MR-PRESSO method as the main method for causal inference and found that genetic susceptibility of insomnia is associated with an increased risk of T2 DM.Combined with the findings of the previous observational studies,this study provides evidence for the causal relationship between insomnia and T2 DM,but we found no evidence of a causal relationship between the other sleep related traits and T2 DM.