Estimating population structure using epigenome-wide methylation data.

Area of Science:

• Genetics
• Epigenetics
• Bioinformatics

Background:

• Population stratification in epigenome-wide association studies (EWAS) can lead to inflated results.
• Accurate estimation of population structure is crucial for reliable EWAS findings.

Purpose of the Study:

• To develop and validate Methylation Population Scores (MPSs) that predict genetic principal components (GPCs).
• To assess the utility of MPSs in addressing population stratification in EWAS.

Main Methods:

• Utilized multi-ethnic methylation data from five cohorts (MESA, CARDIA, JHS, ARIC, HCHS/SOL).
• Employed feature selection and two-stage weighted least squares Lasso regression to identify CpG sites predicting GPCs.
• Adjusted for covariates including age, sex, smoking, alcohol use, BMI, and cell type proportions.

Main Results:

• Developed MPSs demonstrated high correlations with corresponding GPCs (e.g., R²=0.99 for MPS1/GPC1).
• MPSs effectively differentiated self-reported racial/ethnic groups, outperforming previously published methods.
• MPSs showed comparable performance to GPCs in mitigating inflation in EWAS.

Conclusions:

• MPSs provide a reliable estimate of population structure, serving as a valuable complement to GPCs, especially when genetic data is absent.
• This supervised learning approach with covariate adjustment captures genetic structure across diverse populations more effectively than unsupervised methods.
• The derived weights for GPCs can be applied to generate MPSs in other independent studies.