Application of Structural Equation Models to Genome-wide Association Analysis

Genome-wise association studies (GWASs) have become popular approaches to identify genetic variants associated with human biological traits. In this study, we applied Structural Equation Models (SEMs) in order to model complex relationships between genetic networks and traits as risk factors. SEMs allow us to achieve a better understanding of biological mechanisms through identifying greater numbers of genes and pathways that are associated with a set of traits and the relationship among them. For efficient SEM analysis for GWASs, we developed a procedure, comprised of four stages. In the first stage, we conducted single-SNP analysis using regression models, where age, sex, and recruited area were included as adjusting covariates. In the second stage, Fisher’ s combination test was conducted for each gene to detect significant genes using p-values obtained from the single-SNP analysis. In the third stage, Fisher’s exact test was adopted to determine which biological pathways were enriched with significant SNPs. Finally, based on a pathway that was associated with the four traits in common, a SEM was fit to model a causal relationship among the genetic factors and traits. We applied our SEM model to GWAS data with four central obesity related traits: suprailiac and subscapular measures for upper body fat, BMI, and hypertension. Study subjects were collected from two Korean cohort regions. After quality control, 327,872 SNPs for 8842 individuals were included in the analysis. After comparing two SEMs, we concluded that suprailiac and subscapular measures may indirectly affect hypertension susceptibility by influencing BMI. In conclusion, our analysis demonstrates that SEMs provide a better understanding of biological mechanisms by identifying greater numbers of genes and pathways.