Alzheimer’s Polygenic Risk Scores Are Not Interchangeable: Evidence from 1,752 Models

Abstract Introduction Polygenic risk scores (PRS) may improve Alzheimer’s disease (AD) risk prediction before symptom onset, yet choosing an appropriate model can be challenging. Methods Using the standardized GenoPred pipeline, 1,752 PRS models (9 algorithms; 584 configurations; 3 genome-wide association studies) were evaluated and stratified by genetic ancestry and APOE diplotype. PRS models were evaluated using 11,200 clinical or autopsy-confirmed AD cases and 19,321 controls age ≥65 from the Alzheimer’s Disease Sequencing Project Release 5. Results PRS results were not consistent across methodologies (Spearman’s ρ: –0.49 to 1), with >95% of individuals having PRS in both the top and bottom risk deciles. Top-performing PRS were effective at stratifying AD risk across ancestries (AFR: P =2.04×10 - 26 ; AMR: P =4.57×10 -21 ; EAS: P =6.21×10 -40 ; EUR: P =7.90×10 -187 ). Discussion PRS parameters should be optimized for each ancestry. Contradictory signals across methodologies underscore the need for carefully choosing suitable PRS methods and fine-tuning algorithmic parameters to ensure accuracy and consistency. Data Availability Access to the ADSP is controlled by The National Institute on Aging Genetics of Alzheimer’s Disease (NIAGADS). All scripts used to analyze the data are freely available at https://github.com/jmillerlab/prs_comparisons .