Regression models as a tool for genome-wide association studies of Environmental Exposures and DNA Methylation

Epigenetic modifications are correlated to environmental factors. Exposure to ambient air pollution may contribute to the development of different diseases such as cancer, cardiovascular diseases, and neurological and metabolic disorders. Looking for the association between DNA methylation and exposure biomarkers may help in the prevention of adverse effects. Association analysis can be carried out through regression modeling. When dealing with the association between DNA methylation and pollutants, the response variable is beta-distributed, and linear regression models are not appropriate when the range is limited to (0, 1). Beta regression models are more suitable for this situation. Methylation levels can also be measured through the M-value statistic and association studies may be performed using classical linear regression models or robust linear regression models in the presence of outliers. An alternative to these models when the variable of interest does not behave linearly in all the predictors is given by a generalized linear model framework that incorporates non-linear terms and interactions. In this paper, we applied these models to a case study constituted of a cohort of healthy people living in regions exposed to different levels of pollution to investigate the association between DNA methylation and cadmium exposure.