The effect of BMI genetic variants includes interaction components that are challenging to detect reliably. We developed Meta-Quantile Regression (MQR), an approach that utilizes quantile regression and meta-regression to infer potential interactions by modeling variations in genetic effects across the sample distribution of quantitative traits. First, the variability in SNP main effects across these BMI percentiles was shown to result from unadjusted interactions and that these differences capture distributional attributes of interacting variables. Secondly, MQR was used to assess variations in the effect of 37 BMI/obesity-associated SNPs in 75,230 adults of European ancestry. The analysis of BMI gene score shows that genetic interactions shape the genetic architecture of BMI when compared to a similar analysis on 125 height-associated SNPs. Thirdly, the computational cost of fitting MQR was improved using unconditional quantile regression approximations based on influence functions. Lastly, the utility of MQR was compared to variance heterogeneity tests for inferring potential interactions using a simulation study. MQR was found to have a higher power of detecting potential interactions with the number of genotype group levels; under asymmetric error distribution, and antagonistic interactions compared to variance heterogeneity tests while maintaining nominal Type I error rates. Notably, applying rank-based inverse-normal transformations to treat skewness was found to inflate Type I error rates for genotypes with main effects. Overall, MQR is a valuable tool for detecting potential interactions based on the heterogeneity of effects across percentiles of quantitative traits.

Suggested readings:

Abadi, Arkan, et al. "Penetrance of polygenic obesity susceptibility loci across the body mass index distribution." The American Journal of Human Genetics 101.6 (2017): 925-938. (https://www.sciencedirect.com/science/article/pii/S0002929717304287)