Personalized genetics is poised to revolutionize healthcare; however, despite advances, genetic prediction and portability of critical variants remains too limited for clinical use. Genetic prediction is error-prone when applied to individuals with genetic ancestries different from discovery cohorts, often predicting disease risk little better than random in non-European samples. We propose understanding an underlying cause of this loss of prediction accuracy by assessing the extent of GxG interaction effects across ancestries. Using an innovative approach, we tackle this statistically challenging problem by 1) modeling effect sizes of meQTLs on different ancestry haplotypes in an admixed African sample from the same population -- thereby controlling for GxE, and 2) using highly heritable genome-wide methylation phenotypes, affording us thousands of observations per participant rather than a single phenotype (e.g. presence of cardiovascular disease). Our study design allows us to assess whether the presence of ancestry-dependent interactions is a common factor in the variability of SNP effect sizes across populations. Outcomes of this grant include: generating a large genome-wide methylation dataset from 500 admixed South Africans, paired with underlying genome-wide DNA variation. We will further estimate the fraction of meQTLs with ancestry-specific effects and thereby comprehensively provide a snapshot of the frequency of GxG interactions in the human genome. These results will motivate investigation of GxG effects in a broader set of biomedical phenotypes and the extent to which they contribute to poor portability of polygenic risk across populations.