Optimization of multi-ancestry polygenic risk score disease prediction models

Area of Science:

• Genetic Epidemiology
• Computational Biology
• Clinical Diagnostics

Background:

• Polygenic risk scores (PRS) offer insights into disease predisposition but require improved accuracy, interpretability, and portability for clinical use.
• Existing PRS models face challenges in broad clinical application despite advancements.

Purpose of the Study:

• To develop and validate enhanced PRS models for disease prediction with improved clinical utility.
• To assess the performance of PRS algorithms across diverse populations and integrate clinical characteristics for enhanced accuracy.

Main Methods:

• Leveraged trans-ancestry genome-wide association study (GWAS) meta-analysis to generate diverse summary statistics for 30 traits.
• Benchmarked six PRS algorithms using UK Biobank and developed an ensemble model validated on eMERGE and PAGE MEC cohorts.
• Integrated clinical characteristics (age, gender, ancestry, risk factors) into PRS models to create disease prediction tools.

Main Results:

• The ensemble PRS model demonstrated superior performance and good calibration across diverse cohorts.
• Incorporating clinical characteristics significantly improved predictive accuracy, with 12 out of 30 models exceeding 80% AUC.
• High diagnostic odds ratios (DOR) were observed for numerous traits, indicating strong predictive value across all ancestry groups.
• The PRS model for coronary artery disease showed significantly higher identification of events compared to rare variant models.

Conclusions:

• Newly developed PRS-based disease prediction models exhibit sufficient accuracy and portability for clinical consideration.
• The integration of PRS with clinical data offers a promising approach for prospective diagnostic tests.
• Considering both polygenic and rare genetic components is crucial for comprehensive clinical risk assessment.