Transferability of polygenic risk score among diverse ancestries

Polygenic risk score (PRS) has recently emerged as a powerful tool for predicting the risk of complex traits by summarizing weighted single nucleotide polymorphisms (SNPs) associated with the trait(s). The data required for PRS construction, that is, SNPs and effect sizes, can be accessed from in-house association results or previously published association results as deposited in the PGS catalogue (https://www.pgscatalog.org/). PRS has been applicated in risk prediction and stratification for cancers, cardiometabolic diseases, and psychiatric disorders.1, 2 Despite the promising prospect of PRS in clinical application, the application of PRS in the general population is limited by ethnic inequality with respect to the training data. As of 2021, about 86% of genome-wide association study (GWAS) participants are European descendants,3 mainly due to the scarcity of comprehensive genotyping data for non-European populations. Therefore, the prediction accuracy across non-European populations, i.e., the transferability of PRS, is often poor due to the difference in allele frequencies and linkage disequilibrium (LD)4 patterns across populations. The insufficient transferability limits global clinical applicability and may increase disparities in PRS implementation among regions. Most recently, Qu et al.5 addressed this limitation in the application of PRS for body mass index (BMI) in non-European populations. In this study, they developed a globally applicable trans-ethnic PRS for BMI from European ancestry GWAS by generating ethnic-specific LD reference panels and combining them into a Bayesian approach to adjust the GWAS summary statistics. The results demonstrated feasibility and effectiveness of generating trans-ethnic PRS predictions for BMI. To further improve the performance of trans-ethnic PRS in complex traits such as BMI, the most straightforward approach is leveraging large-scale cross-ancestry GWAS data as the sample size expands. Other plausible approaches include fine-mapping or tuning the effect sizes of SNPs to compensate for the limitations imposed by small data sizes. Various statistical models have been applied to improve the prediction accuracy of trans-ethnic PRS based on population genetic structure. State-of-the-art software for computing trans-ethnic PRS includes two recently developed Bayesian polygenic modelling methods, PRS-CSx6 and BridgePRS.7 PRS-CSx uses a method named ‘shared continuous shrinkage prior’ to adjust effect size distribution (i.e. diminishing the small effect size to zero and thereby narrowing the candidate SNPs set), and then jointly utilizes GWAS summary statistics, population-specific allele frequencies, and LD patterns to generate an ultimate PRS through best linear combination. It's a flexible model for cross-ancestry PRS construction considering diverse effect sizes and LD patterns. Nonetheless, PRS-CSx is developed based on an ideal assumption that the causal variants are shared across ancestries and included in PRS construction, which might lack sufficient power in situations where the causal variants are either absent or difficult to identify. To address this issue, BridgePRS applies zero-centred Gaussian prior distribution to adjust effect sizes, focusing on estimating the effect sizes more precisely rather than excluding the SNPs with smaller effect sizes. This method is proven efficient for applying European GWAS summary statistics on non-Europeans. Functional genomics could also be integrated to fine-tune the weighting of effect sizes. Biologically, causal variants tend to have larger functional impacts than tagging SNPs. As early as 2016, the two-dimensional PRS (2D PRS), incorporating external annotations in weighting and categorizing variants, achieved 25%–50% efficiency improvement compared with standard PRS for several diseases.8 Recently, more cell-type-specific regulatory annotations by the IMPACT framework successfully prioritize shared functional variants among populations and increase the transferability of PRS from Europeans to East Asians.9 There are still challenges to overcome in the development and implementation of trans-ethnic PRS. These challenges include ensuring data quality, addressing ethical considerations, and reconciling potential contradictions between self-identified race/ethnicity and genetically inferred ancestry. Further replication studies in independent cohorts are necessary to refine and validate trans-ethnic PRS across a diverse spectrum of complex traits and diseases. Expediting the release of multi-ancestry genomic data and summary statistics is also needed to support the trans-ethnic PRS construction for researchers and clinicians. Moreover, through incorporating additional information, such as multi-omics data, rare variants with large effect sizes, and gene-environment interaction, trans-ethnic PRS holds promising prospects for promiting personalized medicine and public health interventions. The authors have nothing to report. The authors declare no conflict of interest. Not applicable. Data sharing is not applicable to this article as no new data were created or analyzed in this study.