DNA Methylation Markers for Pregnancy Hypertension via Machine Learning Methods

Area of Science:

• Obstetrics and Gynecology
• Epigenetics
• Genomics

Background:

• Hypertensive disorders of pregnancy (HDPs), including gestational hypertension and preeclampsia, pose significant risks to maternal and fetal health.
• Early identification of HDP risk is crucial for timely intervention and improved outcomes.
• Integrating epigenetic biomarkers with clinical factors offers a promising avenue for enhancing prediction models.

Purpose of the Study:

• To develop a predictive model for identifying pregnant individuals at risk of HDPs in the first trimester.
• To investigate the utility of differentially methylated regions (DMRs) as epigenetic biomarkers for HDP prediction.
• To combine epigenetic data with established clinical risk factors for a comprehensive risk assessment.

Main Methods:

• A 2-stage nested case-control study involving 618 pregnant women in China.
• Analysis of peripheral blood samples from the first trimester to assess methylation levels of DMRs.
• Utilized machine learning models (lasso regression, XGBoost, random forest, Shapley Additive Explanations) to identify significant DMRs.
• Constructed prediction models using logistic regression, integrating selected DMRs and clinical factors.

Main Results:

• Identified 52 differential DMRs in the discovery stage and validated 12 DMRs in the second stage.
• Selected three key DMRs located in the CTSA, HMGB1, and miR1908/FADS2 genes for the prediction model.
• The combined model achieved an AUC of 0.863 in the training set and 0.757 in the test set, demonstrating predictive capability.

Conclusions:

• The identified DMRs in peripheral blood show potential as early epigenetic biomarkers for HDP risk.
• This approach offers a novel strategy for early identification of high-risk pregnancies.
• The findings contribute to understanding the epigenetic underpinnings of HDPs.