alternate vs. null models, respectively. Specifically, the BFs were calculated between the full model and models with environmental variables removed to identify which SD-related traits are most relevant for the gene interactions observed. Additionally, we also estimated the fraction of genetic variance explained by multivariate SD-gene interactions. These analyses were conducted separately in each major genetically-determined ancestry group (i.e., African-Americans and European-Americans). We focused our analysis on ancestry-specific analyses only, because the trans-ancestry meta-analysis did not provide findings surviving multiple testing correction due to the fact that the limited sample size of the cohorts investigated was not powerful enough to overcome the heterogeneity due to the different genetic structure (i.e., allele frequency and linkage disequilibrium, LD) of the ancestry groups investigated. The information regarding SI was adjusted for age, sex, genotyping array, and the top 10 PCs, and the residuals obtained were entered as phenotypic outcomes into StructLMM.
