Orienting the causal relationship between imprecisely measured traits using GWAS summary data

Summary

This methodological paper identifies a critical flaw in mediation-based causal inference: measurement error in phenotypes can systematically bias estimates of causal direction, with the problem worsening rather than improving as sample sizes increase. The authors propose an extension to Mendelian randomisation that uses only genome-wide association study summary data and demonstrates greater robustness to measurement error and unmeasured confounding. Application to DNA methylation and gene expression illustrates both the utility and remaining sensitivity of the approach to systematic measurement differences and horizontal pleiotropy.

UK applicability

This methodological advance is relevant to UK biobank and genetics research communities seeking to establish causal relationships between biological traits. The emphasis on triangulating multiple analytical approaches and conducting sensitivity analyses aligns with best practice standards for causal inference in UK epidemiological and genomic studies.

Key measures

Causal direction inference using Mendelian randomisation; bias in causal inference test (CIT) under measurement error; DNA methylation and gene expression associations

Outcomes reported

The study demonstrated that measurement error in phenotypes can cause mediation-based causal inference tests to infer incorrect causal direction, with larger sample sizes paradoxically increasing confidence in wrong conclusions. The authors applied their extended Mendelian randomisation method to infer causal direction between DNA methylation and gene expression, finding DNA methylation generally acts as the causal factor, though results were susceptible to measurement error bias and horizontal pleiotropy.

Topic tags