Graphical analysis for phenome-wide causal discovery in genotyped population-scale biobanks

Abstract

Causal inference via Mendelian randomization requires making strong assumptions about horizontal pleiotropy, where genetic instruments are connected to the outcome not only through the exposure. Here, we present causal Graphical Analysis Using Genetics (cGAUGE), a pipeline that overcomes these limitations using instrument filters with provable properties. This is achievable by identifying conditional independencies while examining multiple traits. cGAUGE also uses ExSep (Exposure-based Separation), a novel test for the existence of causal pathways that does not require selecting instruments. In simulated data we illustrate how cGAUGE can reduce the empirical false discovery rate by up to 30%, while retaining the majority of true discoveries. On 96 complex traits from 337,198 subjects from the UK Biobank, our results cover expected causal links and many new ones that were previously suggested by correlation-based observational studies. Notably, we identify multiple risk factors for cardiovascular disease, including red blood cell distribution width.

MRSL: A phenome-wide causal discovery algorithm based on GWAS summary data

“…We varied across the number of SNPs g 1 and g 2 as 5, 10, 20, 30, 40 and 50, respectively. We compared MRSL with eight published methods: BIMMER [28], cGAUGE based on IVW, MR Egger and MR PRESSO [25], HC algorithm incorporating genetic anchors [23] (based on genetic risk score or the most significant SNP) and MRPC algorithm [24] (based on genetic risk score or the most significant SNP). Details of data generation are shown in Methods section.…”

“…In practice, there may be effects of different magnitude between traits, thus we consider β follows uniform distribution with four parameter settings: U(0,0.25), U(0.25,0.5), U(0. We compared our method with eight published methods: BIMMER [28], cGAUGE based on IVW, MR Egger and MR PRESSO [25], HC algorithm incorporating genetic anchors [23] (based on genetic risk score or the most significant SNP) and MRPC algorithm [24] (based on genetic risk score or the most significant SNP). BIMMER can be implemented using GWAS summary data whereas other seven methods need individual genetic and phenotypic data.…”

“…IVs for the regression (3) above must satisfy the Assumption 4. This kind of IV can be obtained by statistical filtering criteria (see application examples) or ImpIV filter and ExSep test [25].…”

“…Actually, this method only uses the information of genetic variants but not the idea of MR. Afterwards, David et al [25] presented a pipeline, named causal Graphical Analysis Using Genetics (cGAUGE), using IV filters with provable properties to perform univariable MR (UVMR) [26][27] then to obtained a causal graph. This algorithm allows the unobserved confounders among all the variables and requires individual genetic and phenotypic data.…”

Genetic estimates of correlation and causality between blood-based biomarkers and psychiatric disorders

“…Despite this limitation, the existence of genetic correlation between traits is still informative as identifying genes, which affect both psychiatric and biochemical traits, and further insight into the mechanisms would likely refine our understanding of both traits. Moreover, the genetically informed causal inference approaches we implement in this study are subject to limitations regarding the data they are performed with and any biases therein, including potential effects of population stratification ( 50 ), selection bias ( 51 ), and the assumption of acyclicity, which refers to the lack of feedback loops between the exposure and outcome ( 52 ). The LCV model is also fixed to be bivariate in nature, and, thus, the effects of multiple meditators cannot be taken into account.…”