MRC Integrative Epidemiology Unit (IEU) Seminar Series

Title: Latent Heritable Confounder (LHC) model to estimate bi-directional causal effects from GWAS summary statistics

Abstract: 

Mendelian Randomisation (MR) can be used to estimate the causal impact of risk factors on common diseases that represent major public health burden. The InSIDE assumption of most MR methods is violated when genetic instruments affect the outcome not only through the exposure but possibly through a heritable confounder of the exposure-outcome relationship. For this reason, we used a structural equation model corresponding to a causal graph including such unobserved confounder and derived the analytical formula for the resulting variance-covariance structure of the observed joint association summary statistics. The resulting likelihood function was maximized in order to estimate the underlying parameters, such as (direct and indirect) heritability and bi-directional causal effects while exploiting genome-wide GWAS summary statistics for the exposure and the outcome. We compared our causal estimate against those of standard MR methods (Egger, IVW, median- and mode-based estimators) under various simulation scenarios and found that our method has more power, lower bias and root-mean square error especially in the presence of reverse causality and/or strong genetic confounder and lower sample size. Extending the model to incorporate linkage disequilibrium score, sample overlap and population stratification made it applicable to summary statistics for heritable human traits. Using data from the UK Biobank, we observed that while classical MR showed strong bi-directional causal relationship between educational attainment (EDU) and BMI (EDU -> BMI: -0.39, BMI -> EDU: -0.18), our method revealed stronger EDU -> BMI effect (b=-0.69 (p < 1e-12)), but non-significant reverse causality (p > 0.05) and revealed the existence of a genetic confounder (explaining >3% of BMI variation) with opposite effect on BMI and education.

Biography: 

Zoltán Kutalik, PhD is a statistical geneticist, associate professor at the University of Lausanne. In parallel, he is an honorary senior lecturer at the University of Exeter. His main research interest lies in developing statistical methods integrating various omics data in order to better understand the genetic architecture of complex human diseases. He is council member of the Swiss Institute of Bioinformatics (SIB), scientific programme committee member of the European Society of Human Genetics and evaluation committee member of the Swiss National Science Foundation (SNSF). Zoltan is an editorial board member of Human Molecular Genetics, EJHG, and Frontiers in Genetics. He won the Early Career Bioinformatician Award of the SIB, the Investigator-in-training award of the Faculty of Biology and Medicine and shared the Leenaards Prize in 2013. He published over 170 peer-reviewed articles (>27,000 citations, h-index of 69) in international scientific journals, including Nature, Nature Biotechnology, Nature Genetics and Nature Communications. His research has been financed by the SNSF, the SIB, the SystemsX.ch and the Leenaards Foundation.

All welcome