Machine learning models for blood pressure phenotypes combining multiple polygenic risk scores.

Area of Science:

• Genetics and Genomics
• Cardiovascular Disease Research
• Machine Learning Applications

Background:

• Accurate prediction of systolic and diastolic blood pressure (SBP, DBP) is crucial for cardiovascular disease prevention.
• Traditional prediction models often rely on demographic and clinical factors, with limited ability to capture genetic predispositions.
• Polygenic risk scores (PRS) offer a way to quantify an individual's genetic susceptibility to complex traits like hypertension.

Approach:

• Developed non-linear machine learning (ML) ensemble models for SBP and DBP prediction.
• Compared linear versus non-linear models at baseline (demographic/clinical data only) and genetic (including PRS) levels.
• Evaluated the performance improvement from incorporating multiple PRS derived from large genome-wide association studies (GWAS).

Key Points:

• Non-linear baseline models significantly improved prediction accuracy (PVE) over linear models for both SBP and DBP.
• Including seven PRS in the genetic model showed modest improvement compared to a single PRS.
• Adding 14 additional PRS substantially increased prediction accuracy, particularly for non-White populations.

Conclusions:

• Non-linear ML models enhance blood pressure prediction by integrating demographic, clinical, and genetic data.
• Multiple PRS collectively offer greater predictive power than single PRS, with differential benefits across racial/ethnic groups.
• These findings support the utility of advanced ML and PRS in personalized hypertension risk assessment.