PGS-server: accuracy, robustness and transferability of polygenic score methods for biobank scale studies

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Polygenic scores (PGS) are crucial for genetic prediction of complex traits and diseases, advancing precision medicine.
• Despite numerous PGS methods, comprehensive comparative studies evaluating their effectiveness are lacking.
• This knowledge gap hinders optimal application of PGS in research and clinical settings.

Purpose of the Study:

• To conduct a comprehensive comparison of 12 different polygenic score (PGS) methods.
• To evaluate PGS methods based on prediction accuracy, computational scalability, robustness, and transferability.
• To provide practical guidelines for selecting appropriate PGS methods and to introduce a novel aggregation strategy.

Main Methods:

• Internal evaluations of 12 PGS methods on 50 UK Biobank traits (25 quantitative, 25 binary).
• External evaluations using summary statistics from multiple genome-wide association studies (GWAS), including cross-study and cross-ancestry analyses.
• Assessment of prediction accuracy, computational performance, robustness, and transferability across diverse datasets and ancestries.

Main Results:

• Comparative analysis identified strengths and weaknesses of different PGS methods regarding prediction accuracy and scalability.
• Demonstrated the effectiveness of a simple aggregation strategy for improving PGS prediction performance and stability.
• Developed a user-friendly PGS webserver (http://www.pgs-server.com/) for direct application of PGS methods.

Conclusions:

• The study provides essential guidelines for selecting PGS methods, aiding researchers and practitioners.
• The proposed aggregation strategy enhances the reliability and predictive power of polygenic scores.
• The developed PGS webserver facilitates wider adoption and routine application of PGS in genetic research and precision medicine.