Evaluating polygenic risk score prediction performance for Alzheimer's disease in a population-based Hispanic cohort using single- and multi-ancestry models.

Background: The Polygenic risk score (PRS) is effective in predicting Alzheimer's Disease (AD) risk among Europeans but remains understudied in Hispanics. Genome-wide association studies (GWAS) based on multiple ancestries can improve PRS prediction. We used GWAS data from the largest available African, European, and Hispanic populations and performed PRS analyses using novel methodologies to evaluate the performance of single- and multi-ancestry PRS models in predicting AD risk among Hispanic population.

Methods: Prediction performance of Apolipoprotein-E (), single-ancestry PRS, and multi-ancestry PRS derived from GWAS-focused and method-focused approaches to clinical AD, incident AD, and cognition and were evaluated in 2961 Hispanic people from two large studies. The GWAS-focused approach constructs PRS based on multi-ancestry GWAS, while the method-focused approach uses novel multi-ancestry PRS methods, integrating GWAS summary statistics across ancestries. Ten repetitions of 5-fold cross-validation were used. In a subset, plasma biomarker data were used in a tuning-validation split to examine PRS performance in predicting single and combined biomarkers.

Findings: The multi-ancestry PRS excluding , constructed using the method-focused approach, outperformed both single-ancestry and multi-ancestry PRSs from the GWAS-focused approach. The best method-focused PRS, incorporating summary statistics from GWASs of African, European, and Hispanic populations, explained up to 1.6%, 3.9%, and 1.7% of the variance in clinical AD, incident AD, and cognition, respectively-comparable to, or even higher than, the variance explained by the . Similar findings were observed in biomarker analyses. accounted for more variation in plasma P-tau levels and PRS explained more variation in Aβ levels.

Interpretation: Integrating novel multi-ancestry PRS methods (e.g., PROSPER/PRS-CSx) with GWAS across ancestries enhances prediction accuracy for AD risk among Hispanic population. and PRS may point to different biological aspects of AD.

Funding: National Institutes of Health R01 AG072474, RF1 AG066107, 5R37AG015473, RF1AG015473, R56AG051876, R01 AG067501, and UL1TR001873.