Fig. 2

Epigenetic inference of alcohol consumption from blood based on Liu et al. biomarkers and models. Prediction accuracy for alcohol consumption expressed as Area Under the Curve (AUC) for A heavy drinkers vs. non-drinkers and B heavy drinkers vs. light drinkers using the CpG marker sets from Liu et al. [12]. Data from participants who do not fit the inferred categories were excluded from the respective prediction models following the approach used by Liu et al. ‘Internal Validation’: Mean AUC and SD from internal validation using ten-fold cross-validation in our model building dataset. ‘External Validation’: AUCs from external validation by applying our models trained in the model building dataset to independent data from three external validation cohorts (Rotterdam Study, N = 648; SHIP-Trend, N = 433; and TwinsUK, N = 713 and N = 442). Based on interview or self-reported information, non-drinkers were defined as participants with no alcohol consumption; light drinkers with an alcohol consumption of 0 < g per day ⩽28 in men and 0 < g per day ⩽14 in women; and heavy drinkers with an alcohol consumption of ⩾42 g per day in men and ⩾28 g per day in women. Abbreviations: RS- The Rotterdam Study; SHIP- Study of Health in Pomerania-Trend cohort; TwinsUK- The TwinsUK Study; TwinsUK2- Subset of the TwinsUK Study