Search

Great efforts are being made to develop advanced polygenic risk scores (PRS) to improve the prediction of complex traits and diseases. However, most existing PRS are primarily trained on European ancestry populations, limiting their transferability to non-European populations. In this article, we propose a novel method for generating multi-ancestry Polygenic Risk scOres based on enSemble of PEnalized Regression models (PROSPER). PROSPER integrates genome-wide association studies (GWAS) summary statistics from diverse populations to develop ancestry-specific PRS with improved predictive power for minority populations. The method uses a combination of ℒ1(lasso) and ℒ2(ridge) penalty functions, a parsimonious specification of the penalty parameters across populations, and an ensemble step to combine PRS generated across different penalty parameters. We evaluate the performance of PROSPER and other existing methods on large-scale simulated and real datasets, including those from 23andMe Inc., the Global Lipids Genetics Consortium, and All of Us. Results show that PROSPER can substantially improve multi-ancestry polygenic prediction compared to alternative methods across a wide variety of genetic architectures. In real data analyses, for example, PROSPER increased out-of-sample prediction R2for continuous traits by an average of 70% compared to a state-of-the-art Bayesian method (PRS-CSx) in the African ancestry population. Further, PROSPER is computationally highly scalable for the analysis of large SNP contents and many diverse populations.

Biological experiments often involve hypothesis testing at the scale of thousands to millions of tests. Alleviating the multiple testing burden has been a goal of many methods designed to boost test power by focusing tests on the alternative hypotheses most likely to be true. Very often, these methods either explicitly or implicitly make use of prior probabilities that bias significance for favored sets thought to be enriched for significant finding. Nevertheless, most genomics experiments, and in particular genome-wide association studies (GWAS), still use traditional univariate tests rather than more sophisticated approaches. Here we use GWAS to demonstrate why unbiased tests remain in favor. We calculate test power assuming perfect knowledge of a prior distribution and then derive the population size increase required to provided the same boost without a prior. We show that population size is exponentially more important than prior, providing a rigorous explanation for the observed avoidance of prior-based methods. Author summary Biological experiments often test thousands to millions of hypotheses. Gene-based tests for human RNA-Seq data, for example, involve approximately 20,000; genome-wide association studies (GWAS) involve about 1 million effective tests. The conventional approach is to perform individual tests and then apply a Bonferroni correction to account for multiple testing. This approach implies a single-test p-value of 2.5 × 10 −6 for RNA-Seq experiments, and a p-value of 5 × 10 −8 for GWAS, to control the false-positive rate at a conventional value of 0.05. Many methods have been proposed to alleviate the multiple-testing burden by incorporating a prior probability that boosts the significance for a subset of candidate genes or variants. At the extreme limit, only the candidate set is tested, corresponding to a decreased multiple testing burden. Despite decades of methods development, prior-based tests have not been generally used. Here we compare the power increase possible with a prior with the increase possible with a much simpler strategy of increasing a study size. We show that increasing the population size is exponentially more valuable than increasing the strength of prior, even when the true prior is known exactly. These results provide a rigorous explanation for the continued use of simple, robust methods rather than more sophisticated approaches.

Great efforts are being made to develop advanced polygenic risk scores (PRS) to improve the prediction of complex traits and diseases. However, most existing PRS are primarily trained on European ancestry populations, limiting their transferability to non-European populations. In this article, we propose a novel method for generating multi-ancestry Polygenic Risk scOres based on enSemble of PEnalized Regression models (PROSPER). PROSPER integrates genome-wide association studies (GWAS) summary statistics from diverse populations to develop ancestry-specific PRS with improved predictive power for minority populations. The method uses a combination of ℒ 1 (lasso) and ℒ 2 (ridge) penalty functions, a parsimonious specification of the penalty parameters across populations, and an ensemble step to combine PRS generated across different penalty parameters. We evaluate the performance of PROSPER and other existing methods on large-scale simulated and real datasets, including those from 23andMe Inc., the Global Lipids Genetics Consortium, and All of Us. Results show that PROSPER can substantially improve multi-ancestry polygenic prediction compared to alternative methods across a wide variety of genetic architectures. In real data analyses, for example, PROSPER increased out-of-sample prediction R 2 for continuous traits by an average of 70% compared to a state-of-the-art Bayesian method (PRS-CSx) in the African ancestry population. Further, PROSPER is computationally highly scalable for the analysis of large SNP contents and many diverse populations., Competing Interests: Competing Interests Statement J.Zhan, J.O., Y.J. and B.L.K. are employed by and hold stock or stock options in 23andMe, Inc. The remaining authors declare no competing interests.

Polygenic risk scores (PRS) are now showing promising predictive performance on a wide variety of complex traits and diseases, but there exists a substantial performance gap across different populations. We propose MUSSEL, a method for ancestry-specific polygenic prediction that borrows information in the summary statistics from genome-wide association studies (GWAS) across multiple ancestry groups. MUSSEL conducts Bayesian hierarchical modeling under a MUltivariate Spike-and-Slab model for effect-size distribution and incorporates an Ensemble Learning step using super learner to combine information across different tuning parameter settings and ancestry groups. In our simulation studies and data analyses of 16 traits across four distinct studies, totaling 5.7 million participants with a substantial ancestral diversity, MUSSEL shows promising performance compared to alternatives. The method, for example, has an average gain in prediction R 2 across 11 continuous traits of 40.2% and 49.3% compared to PRS-CSx and CT-SLEB, respectively, in the African Ancestry population. The best-performing method, however, varies by GWAS sample size, target ancestry, underlying trait architecture, and the choice of reference samples for LD estimation, and thus ultimately, a combination of methods may be needed to generate the most robust PRS across diverse populations., Competing Interests: Declaration of interests J.Zhan, Y.J., J.O., and B.L.K. are employed by and hold stock or stock options in 23andMe, Inc.