bioRxiv2021
It is of great interest and potential to discover causal relationships between pairs of exposures and outcomes using genetic variants as instrumental variables (IVs) to deal with hidden confounding in observational studies. Two most popular approaches are Mendelian randomization (MR), which usually use independent genetic variants/SNPs across the genome, and transcriptome-wide association studies (TWAS) using cis-SNPs local to a gene, as IVs. In spite of their many promising applications, both approaches face a major challenge: the validity of their causal conclusions depends on three critical assumptions on valid IVs, which however may not hold in practice. The most likely as well as challenging situation is due to the wide-spread horizontal pleiotropy, leading to two of three IV assumptions being violated and thus to biased statistical inference. More generally, wed like to conduct a goodness-of-fit (GOF) test to check the model being used. Although some methods have been proposed as being robust to various degrees to the violation of some modeling assumptions, they often give different and even conflicting results due to their own modeling assumptions and possibly lower statistical efficiency, imposing difficulties to the practitioner in choosing and interpreting varying results across different methods. Hence, it would help to directly test whether any assumption is violated or not. In particular, there is a lack of such tests for TWAS. We propose a new and general GOF test, called TEDE (TEsting Direct Effects), applicable to both correlated and independent SNPs/IVs (as commonly used in TWAS and MR respectively). Through simulation studies and real data examples, we demonstrate high statistical power and advantages of our new method, while confirming the frequent violation of modeling (including IV) assumptions in practice and thus the importance of model checking by applying such a test in MR/TWAS analysis. Author SummaryWith the increasing availability of large-scale GWAS summary data of various complex traits/diseases and software packages, it has become convenient and popular to apply Mendelian randomization (MR) and transcriptome-wide association studies (TWAS), using genetic variants as instrumental variables (IVs), to address fundamental and significant questions by unraveling causal relationships between complex or molecular traits such as gene expression and other complex traits. However, the validity of such causal conclusions critically depends on the validity of the model being used, including three key IV assumptions. In particular, with the wide-spread horizontal pleiotropy of genetic variants, two of the three IV assumptions may be violated, leading to biased inference from MR and TWAS. This issue may become more severe as more trait-associated genetic variants are used as IVs to increase the power of MR and TWAS. Although there are some methods to check the modeling assumptions for MR with independent genetic variants as IVs, there is barely any powerful one for TWAS (or more generally for MR and similar methods) with correlated SNPs as IVs. We propose such a powerful method applicable to both MR and TWAS with local or genome-wide, possibly correlated, SNPs as IVs, demonstrating its higher statistical power than several commonly used methods, while confirming the frequent violation of modeling/IV assumptions in TWAS with our example GWAS data of schizophrenia, Alzheimers disease and blood lipids. An important conclusion is that in practice it is necessary to conduct model checking in MR and TWAS, and our proposed method is expected to be useful for such a task.