Constructing a Literature-Derived Database for Benchmarking Polygenic Risk Score Construction Methods with Spectral...

Area of Science:

• Genetics
• Bioinformatics
• Computational Biology

Background:

• Polygenic risk scores (PRSs) are vital for predicting and stratifying genetic risk in human diseases.
• Numerous methods for PRS construction have been developed, necessitating comprehensive benchmarking.
• Existing evidence on PRS method performance is fragmented, lacking a unified assessment.

Purpose of the Study:

• To systematically benchmark and rank various polygenic risk score (PRS) construction methods.
• To create a comprehensive database synthesizing published PRS method comparisons.
• To provide guidance for selecting appropriate PRS methods in genetic research and clinical applications.

Main Methods:

• Constructed a database of PRS method benchmarking results from 2009 to 2025.
• Applied a spectral ranking inference framework with uncertainty quantification.
• Ranked 14 PRS methods using data from original development and application studies.

Main Results:

• Identified LDpred2 and AnnoPred as consistently top-ranked PRS methods, and C+T as the lowest-ranked.
• Observed moderate variability in the relative ordering of most PRS methods across different sources.
• Generated phenotype-specific rankings revealing method-specific strengths and robustness.

Conclusions:

• The study provides a systematic ranking and curated database of PRS methods.
• The findings offer a dynamic reference to guide the selection of PRS methods for future applications.
• Phenotype-specific rankings enhance the practical utility of the PRS method assessment.