Fig. 1: Summary of gene-by-environment (GxE) analysis conducted with MonsterLM.
Initial simulation studies were conducted to verify the properties of MonsterLM. Simulated outcomes with known values for variance explained were regressed under varying scenarios and model specifications to ensure robust estimation (blue panel). Real outcome analyses were conducted with UK Biobank data (grey panels). Genome-wide SNP heritability estimates with and without waist-hip-ratio (WHR) interactions revealed significant interaction effects for 8 of 13 outcomes and were further assessed with a directionality of effects and stratification analysis (bottom left panel). MonsterLM properties were further explored recovering genotype and interaction variance explained through partitioning SNPs based on genotype and interaction univariate regressions (bottom middle panel). Lastly, sequential incorporation of subsets of SNPs with significant interaction associations derived from univariate interaction regressions of the genotype SNPs on their respective outcomes revealed modest improvements of polygenic scores in one of the eight outcomes tested (bottom right panel). Genotype Matrices: Block partitioning schematic. Base and Collider Models: green circles are outcome variables; red circles are predictor variables; red squares are colliders; blue circles are confounders; green arrows are causal associations; grey arrows are unobserved causal associations. R2G: Genetic variance; R2GE: GxE variance; \({P}_{G}\): univariate genetic association SNPs; \({P}_{{GE}}\): univariate interaction association SNPs.
