SDPRX: A statistical method for cross-population prediction of complex traits

Real-time dynamic polygenic prediction for streaming data

Polygenic risk scores (PRSs) are promising tools for advancing precision medicine. However, existing PRS construction methods rely on static summary statistics derived from genome-wide association studies (GWASs), which are often updated at lengthy intervals. As genetic data and health outcomes are continuously being generated at an ever-increasing pace, the current PRS training and deployment paradigm is suboptimal in maximizing the prediction accuracy of PRSs for incoming patients in healthcare settings. Here, we introduce real-time PRS-CS (rtPRS-CS), which enables online, dynamic refinement and calibration of PRS as each new sample is collected, without the need to perform intermediate GWASs. Through extensive simulation studies, we evaluate the performance of rtPRS-CS across various genetic architectures and training sample sizes. Leveraging quantitative traits from the Mass General Brigham Biobank and UK Biobank, we show that rtPRS-CS can integrate massive streaming data to enhance PRS prediction over time. We further apply rtPRS-CS to 22 schizophrenia cohorts in 7 Asian regions, demonstrating the clinical utility of rtPRS-CS in dynamically predicting and stratifying disease risk across diverse genetic ancestries.

Stratifying Lung Adenocarcinoma Risk with Multi-ancestry Polygenic Risk Scores in East Asian Never-Smokers

Background

Lung adenocarcinoma (LUAD) among never-smokers is a public health burden especially prevalent in East Asian (EAS) women. Polygenic risk scores (PRSs), which quanefy geneec suscepebility, are promising for straefying risk, yet have mainly been developed in European (EUR) populaeons. We developed and validated single-and mule-ancestry PRSs for LUAD in EAS never-smokers, using the largest available genome-wide associaeon study (GWAS) dataset.

Methods

We used GWAS summary staesecs from both EAS (8,002 cases; 20,782 controls) and EUR (2,058 cases; 5,575 controls) populaeons, as well as independent EAS individual level data. We evaluated several PRSs approaches: a single-ancestry PRS using 25 variants that reached genome-wide significance (PRS-25), a genome-wide Bayesian based approach (LDpred2), and a mule-ancestry approach that models geneec correlaeons across ancestries (CT-SLEB). PRS performance was evaluated based on the associaeon with LUAD and AUC values. We then esemated the lifeeme absolute risk of LUAD (age 30-80) and projected the AUC at different sample sizes using EAS-derived effect-size distribueon and heritability esemates.

Findings

The CT-SLEB PRS showed a strong associaeon with LUAD risk (odds raeo=1.71, 95% confidence interval (CI): 1.61, 1.82) with an AUC of 0.640 (95% CI: 0.629, 0.653). Individuals in the 95 th percenele of the PRS had an esemated 6.69% lifeeme absolute risk of LUAD. Comparison of LUAD risk between individuals in the highest and lowest 20% PRS quaneles revealed a 3.92-fold increase. Projeceon analyses indicated that achieving an AUC of 0.70, which approaches the maximized prediceon poteneal of the PRS given the esemated geneec variance, would require a future study encompassing 55,000 EAS LUAD cases with a 1:10 case-control raeo.

Interpretations

Our study underscores the poteneal of mule-ancestry PRS approaches to enhance LUAD risk straeficaeon in never-smokers, parecularly in EAS populaeons, and highlights the necessary scale of future research to uncover the geneec underpinnings of LUAD.

Genomic Insights for Personalized Care: Motivating At-Risk Individuals Toward Evidence-Based Health Practices

Background

Lung cancer and tobacco use pose significant global health challenges, necessitating a comprehensive translational roadmap for improved prevention strategies. Polygenic risk scores (PRSs) are powerful tools for patient risk stratification but have not yet been widely used in primary care for lung cancer, particularly in diverse patient populations.

Methods

We propose the GREAT care paradigm, which employs PRSs to stratify disease risk and personalize interventions. We developed PRSs using large-scale multi-ancestry genome-wide association studies and standardized PRS distributions across all ancestries. We applied our PRSs to 796 individuals from the GISC Trial, 350,154 from UK Biobank (UKBB), and 210,826 from All of Us Research Program (AoU), totaling 561,776 individuals of diverse ancestry.

Results

Significant odds ratios (ORs) for lung cancer and difficulty quitting smoking were observed in both UKBB and AoU. For lung cancer, the ORs for individuals in the highest risk group (top 20% versus bottom 20%) were 1.85 (95% CI: 1.58 - 2.18) in UKBB and 2.39 (95% CI: 1.93 - 2.97) in AoU. For difficulty quitting smoking, the ORs (top 33% versus bottom 33%) were 1.36 (95% CI: 1.32 - 1.41) in UKBB and 1.32 (95% CI: 1.28 - 1.36) in AoU.

Conclusion

Our PRS-based intervention model leverages large-scale genetic data for robust risk assessment across populations. This model will be evaluated in two cluster-randomized clinical trials aimed at motivating health behavior changes in high-risk patients of diverse ancestry. This pioneering approach integrates genomic insights into primary care, promising improved outcomes in cancer prevention and tobacco treatment.

Optimizing clinico-genomic disease prediction across ancestries: a machine learning strategy with Pareto improvement

BACKGROUND

Accurate prediction of an individual's predisposition to diseases is vital for preventive medicine and early intervention. Various statistical and machine learning models have been developed for disease prediction using clinico-genomic data. However, the accuracy of clinico-genomic prediction of diseases may vary significantly across ancestry groups due to their unequal representation in clinical genomic datasets.

METHODS

We introduced a deep transfer learning approach to improve the performance of clinico-genomic prediction models for data-disadvantaged ancestry groups. We conducted machine learning experiments on multi-ancestral genomic datasets of lung cancer, prostate cancer, and Alzheimer's disease, as well as on synthetic datasets with built-in data inequality and distribution shifts across ancestry groups.

RESULTS

Deep transfer learning significantly improved disease prediction accuracy for data-disadvantaged populations in our multi-ancestral machine learning experiments. In contrast, transfer learning based on linear frameworks did not achieve comparable improvements for these data-disadvantaged populations.

CONCLUSIONS

This study shows that deep transfer learning can enhance fairness in multi-ancestral machine learning by improving prediction accuracy for data-disadvantaged populations without compromising prediction accuracy for other populations, thus providing a Pareto improvement towards equitable clinico-genomic prediction of diseases.

Evaluating and improving health equity and fairness of polygenic scores

Polygenic scores (PGSs) are quantitative metrics for predicting phenotypic values, such as human height or disease status. Some PGS methods require only summary statistics of a relevant genome-wide association study (GWAS) for their score. One such method is Lassosum, which inherits the model selection advantages of Lasso to select a meaningful subset of the GWAS single-nucleotide polymorphisms as predictors from their association statistics. However, even efficient scores like Lassosum, when derived from European-based GWASs, are poor predictors of phenotype for subjects of non-European ancestry; that is, they have limited portability to other ancestries. To increase the portability of Lassosum, when GWAS information and estimates of linkage disequilibrium are available for both ancestries, we propose Joint-Lassosum (JLS). In the simulation settings we explore, JLS provides more accurate PGSs compared to other methods, especially when measured in terms of fairness. In analyses of UK Biobank data, JLS was computationally more efficient but slightly less accurate than a Bayesian comparator, SDPRX. Like all PGS methods, JLS requires selection of predictors, which are determined by data-driven tuning parameters. We describe a new approach to selecting tuning parameters and note its relevance for model selection for any PGS. We also draw connections to the literature on algorithmic fairness and discuss how JLS can help mitigate fairness-related harms that might result from the use of PGSs in clinical settings. While no PGS method is likely to be universally portable, due to the diversity of human populations and unequal information content of GWASs for different ancestries, JLS is an effective approach for enhancing portability and reducing predictive bias.

Tuning parameters for polygenic risk score methods using GWAS summary statistics from training data

Various polygenic risk scores (PRS) methods have been proposed to combine the estimated effects of single nucleotide polymorphisms (SNPs) to predict genetic risks for common diseases, using data collected from genome-wide association studies (GWAS). Some methods require external individual-level GWAS dataset for parameter tuning, posing privacy and security-related concerns. Leaving out partial data for parameter tuning can also reduce model prediction accuracy. In this article, we propose PRStuning, a method that tunes parameters for different PRS methods using GWAS summary statistics from the training data. PRStuning predicts the PRS performance with different parameters, and then selects the best-performing parameters. Because directly using training data effects tends to overestimate the performance in the testing data, we adopt an empirical Bayes approach to shrinking the predicted performance in accordance with the genetic architecture of the disease. Extensive simulations and real data applications demonstrate PRStuning's accuracy across PRS methods and parameters.

MESuSiE enables scalable and powerful multi-ancestry fine-mapping of causal variants in genome-wide association studies

Fine-mapping in genome-wide association studies attempts to identify causal SNPs from a set of candidate SNPs in a local genomic region of interest and is commonly performed in one genetic ancestry at a time. Here, we present multi-ancestry sum of the single effects model (MESuSiE), a probabilistic multi-ancestry fine-mapping method, to improve the accuracy and resolution of fine-mapping by leveraging association information across ancestries. MESuSiE uses summary statistics as input, accounts for the diverse linkage disequilibrium pattern observed in different ancestries, explicitly models both shared and ancestry-specific causal SNPs, and relies on a variational inference algorithm for scalable computation. We evaluated the performance of MESuSiE through comprehensive simulations and multi-ancestry fine-mapping of four lipid traits with both European and African samples. In the real data, MESuSiE improves fine-mapping resolution by 19.0% to 72.0% compared to existing approaches, is an order of magnitude faster, and captures and categorizes shared and ancestry-specific causal signals with enhanced functional enrichment.

Principles and methods for transferring polygenic risk scores across global populations

Polygenic risk scores (PRSs) summarize the genetic predisposition of a complex human trait or disease and may become a valuable tool for advancing precision medicine. However, PRSs that are developed in populations of predominantly European genetic ancestries can increase health disparities due to poor predictive performance in individuals of diverse and complex genetic ancestries. We describe genetic and modifiable risk factors that limit the transferability of PRSs across populations and review the strengths and weaknesses of existing PRS construction methods for diverse ancestries. Developing PRSs that benefit global populations in research and clinical settings provides an opportunity for innovation and is essential for health equity.

Leveraging genetic correlations and multiple populations to improve genetic risk prediction for non-European populations

The disparity in genetic risk prediction accuracy between European and non-European individuals highlights a critical challenge in health inequality. To bridge this gap, we introduce JointPRS, a novel method that models multiple populations jointly to improve genetic risk predictions for non-European individuals. JointPRS has three key features. First, it encompasses all diverse populations to improve prediction accuracy, rather than relying solely on the target population with a singular auxiliary European group. Second, it autonomously estimates and leverages chromosome-wise cross-population genetic correlations to infer the effect sizes of genetic variants. Lastly, it provides an auto version that has comparable performance to the tuning version to accommodate the situation with no validation dataset. Through extensive simulations and real data applications to 22 quantitative traits and four binary traits in East Asian populations, nine quantitative traits and one binary trait in African populations, and four quantitative traits in South Asian populations, we demonstrate that JointPRS outperforms state-of-art methods, improving the prediction accuracy for both quantitative and binary traits in non-European populations.

Evaluating and Improving Health Equity and Fairness of Polygenic Scores

Polygenic scores (PGS) are quantitative metrics for predicting phenotypic values, such as human height or disease status. Some PGS methods require only summary statistics of a relevant genome-wide association study (GWAS) for their score. One such method is Lassosum, which inherits the model selection advantages of Lasso to select a meaningful subset of the GWAS single nucleotide polymorphisms as predictors from their association statistics. However, even efficient scores like Lassosum, when derived from European-based GWAS, are poor predictors of phenotype for subjects of non-European ancestry; that is, they have limited portability to other ancestries. To increase the portability of Lassosum, when GWAS information and estimates of linkage disequilibrium are available for both ancestries, we propose Joint-Lassosum. In the simulation settings we explore, Joint-Lassosum provides more accurate PGS compared with other methods, especially when measured in terms of fairness. Like all PGS methods, Joint-Lassosum requires selection of predictors, which are determined by data-driven tuning parameters. We describe a new approach to selecting tuning parameters and note its relevance for model selection for any PGS. We also draw connections to the literature on algorithmic fairness and discuss how Joint-Lassosum can help mitigate fairness-related harms that might result from the use of PGS scores in clinical settings. While no PGS method is likely to be universally portable, due to the diversity of human populations and unequal information content of GWAS for different ancestries, Joint-Lassosum is an effective approach for enhancing portability and reducing predictive bias.

Improving genetic risk prediction across diverse population by disentangling ancestry representations

Risk prediction models using genetic data have seen increasing traction in genomics. However, most of the polygenic risk models were developed using data from participants with similar (mostly European) ancestry. This can lead to biases in the risk predictors resulting in poor generalization when applied to minority populations and admixed individuals such as African Americans. To address this issue, largely due to the prediction models being biased by the underlying population structure, we propose a deep-learning framework that leverages data from diverse population and disentangles ancestry from the phenotype-relevant information in its representation. The ancestry disentangled representation can be used to build risk predictors that perform better across minority populations. We applied the proposed method to the analysis of Alzheimer's disease genetics. Comparing with standard linear and nonlinear risk prediction methods, the proposed method substantially improves risk prediction in minority populations, including admixed individuals, without needing self-reported ancestry information.

Predicting Breast Cancer Risk for Women Veterans of African Ancestry in the Million Veteran Program

Breast cancer is a leading cause of cancer and, therefore, a major health threat for women in the United States and worldwide. We have seen over the years major advances in breast cancer prevention and care. Breast cancer screening with mammography leads to reduction in breast cancer mortality, and breast cancer prevention treatment with antiestrogens results in reduction in breast cancer incidence. More progress, however, is urgently needed for this common cancer that affects 1 in 11 American women in their lifetime. Not all women have the same breast cancer risk. A personalized approach is highly desirable as women with higher breast cancer risk may benefit from more intense breast cancer screening and/or prevention intervention while lower risk women may be spared with the cost, inconvenience, and emotional burden of these procedures. In addition to age, demographics, family history, lifestyle, and personal health, genetics is an important determinant of an individual's risk for breast cancer. Over the past 10 years, advances in cancer genomics identified multiple common genetic variants from population studies that collectively can contribute significantly to an individual's breast cancer risk. The effects of these genetic variants can be summarized as a "polygenic risk score" (PRS). We are among the first groups to prospectively evaluate the performance of these risk prediction instruments among women veterans of the Million Veteran Program (MVP). A 313-variant PRS (PRS313) predicted incident breast cancer for a prospective cohort of European (EUR) ancestry women veterans with an area under the receiver operating characteristic curve (AUC) of 0.622. The PRS313 performed less well for AFR ancestry however, with an AUC of 0.579. This is not surprising as most genome-wide association studies were conducted in people of European ancestry. This is an important area of health disparity and unmet need. The large population size and diversity of the MVP provide a unique and important opportunity to explore novel approaches to produce accurate and clinically useful genetic risk prediction instruments for minority populations.