Real-time dynamic polygenic prediction for streaming data

Area of Science:

• Genetics
• Bioinformatics
• Precision Medicine

Background:

• Polygenic risk scores (PRSs) are crucial for precision medicine but rely on outdated genome-wide association study (GWAS) data.
• Current PRS methods are static, limiting their predictive accuracy for incoming patients as new data emerges.
• There is a need for dynamic PRS construction that integrates continuously generated genetic and health outcome data.

Purpose of the Study:

• To introduce real-time PRS-CS (rtPRS-CS), a novel method for online, dynamic refinement of PRSs.
• To evaluate the performance of rtPRS-CS in enhancing PRS prediction accuracy using streaming data.
• To demonstrate the clinical utility of rtPRS-CS in diverse populations and for disease risk prediction.

Main Methods:

• Developed rtPRS-CS for online, dynamic PRS construction and standardization with each new sample.
• Conducted extensive simulations to assess rtPRS-CS performance across various genetic architectures and sample sizes.
• Applied rtPRS-CS to quantitative traits from two large biobanks and 22 schizophrenia cohorts across Asian regions.

Main Results:

• rtPRS-CS effectively integrates massive streaming data to improve PRS prediction accuracy over time.
• Simulations confirmed rtPRS-CS's robustness across different genetic architectures and training sample sizes.
• Demonstrated clinical utility of rtPRS-CS in dynamically capturing health status changes and predicting disease risk in diverse ancestries.

Conclusions:

• rtPRS-CS offers a significant advancement over static PRS methods by enabling real-time adaptation.
• The dynamic nature of rtPRS-CS enhances predictive power for precision medicine applications.
• rtPRS-CS shows promise for improving disease risk prediction and clinical management across diverse populations.