Polygenic Indexes Show Up to 76% Reduced Predictive Accuracy in Non-European Ancestries, New Study Reveals

A recent study, "Dissecting the Predictive Accuracy of Polygenic Indexes for Behavioral Phenotypes Across Genetic Ancestries," has highlighted significant limitations in the cross-ancestry portability of polygenic indexes (PGIs). Shared by author Razib Khan, the research, published as a preprint on bioRxiv, systematically analyzes PGI performance for 52 health-related, behavioral, and social phenotypes, confirming substantial reductions in predictive power when applied to non-European populations. The findings underscore a critical challenge in applying genomic insights equitably across diverse global populations.

The study, utilizing data from the UK Biobank and Health and Retirement Study, found that PGIs trained on European genetic ancestries exhibited systematic reductions in predictive power for non-European ancestries. The predictive accuracy was lowest in African ancestries, with a 24% retention (76% loss), followed by East Asian (37% retention) and South Asian (51% retention) genetic ancestries. Biologically proximal traits demonstrated greater portability compared to behavioral and social traits, indicating varying degrees of genetic influence and environmental interaction.

Researchers identified linkage disequilibrium (LD) and allele frequency differences as primary drivers of this accuracy loss, explaining 82% of the reduction in African ancestries. While these factors contributed less significantly in East (34%) and South Asian (25%) ancestries, the overall trend points to inherent genetic architectural differences. The study also explored family-based GWAS PGIs, noting modest improvements in portability for specific traits like BMI in African ancestry, suggesting that some portability gaps may stem from population-specific confounds in standard PGIs.

This research reinforces a well-recognized "portability problem" in genomics, where the historical overrepresentation of European ancestries in genome-wide association studies (GWAS) limits the applicability of derived polygenic scores to other groups. This disparity raises concerns about exacerbating health inequities if such tools are implemented clinically without addressing their differential accuracy. The findings emphasize the urgent need for more ancestrally diverse genomic datasets and refined methodologies to ensure equitable benefits from genetic prediction.

The implications extend to the ethical considerations surrounding the use of PGIs, particularly for complex behavioral traits. As Razib Khan stated in the tweet, the work focuses on "Dissecting the Predictive Accuracy of Polygenic Indexes for Behavioral Phenotypes Across Genetic Ancestries," highlighting the intricate nature of these predictions. Future research efforts are called upon to prioritize inclusive genetic studies and develop robust, multi-ancestry models to bridge these predictive divides and foster more generalized and equitable genomic applications.