Seunggeun Lee
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Seunggeun Lee has a total of 839 co-authors. Here are the top co-authors by number of shared publications:
- Lars G. Fritsche
Department of Biostatistics, University of Michigan School of Public Health, Ann Arbor, USA
23 shared publications - Wei Zhou
Program in Medical and Population Genetics, Broad Institute of MIT and Harvard, Cambridge, USA
19 shared publications - Cristen Willer
Department of Internal Medicine, Division of Cardiology, University of Michigan, Ann Arbor, MI, USA
14 shared publications - Jonas Bille Nielsen
K.G. Jebsen Center for Genetic Epidemiology, Department of Public Health and Nursing, NTNU, Norwegian University of Science and Technology, Trondheim, Norway
13 shared publications Wei Zhou
Psychiatric and Neurodevelopmental Genetics Unit, Center for Genomic Medicine, Massachusetts General Hospital
12 shared publications- Bhramar Mukherjee
Yale University, USA
11 shared publications
Latest Publications
- Open Access
- 26 October 2026
- Published by Frontiers Media SA in Frontiers in Pharmacology
- Vol. 17, 1892708
Integrating whole-exome and genotyping array data for pharmacogenetic typing: a comparative evaluation with whole-genome sequencing in UK Biobank
Introduction: Whole‐genome sequencing (WGS) provides comprehensive coverage of coding, non‐coding, and rare variants for pharmacogenetic star‐allele typing, but cost and computational demands limit its population-scale use. Whole-exome sequencing (WES) and imputed genotyping arrays are more widely available, yet their WGS-referenced performance remains incompletely defined. Methods: We benchmarked star-allele calls from WES, imputed array data, and an integrated call set combining WES and array data against an Aldy-derived WGS analytical reference across 33 pharmacogenes in 1,450 multi-ancestry UK Biobank participants. We also developed PGxIntegrator, an open‐source workflow that harmonizes WES and array data and enables scalable Aldy‐based star‐allele calling. Results: Using WES or imputed array data alone, 25 of 33 genes reached high diplotype concordance with the WGS analytical reference (≥95%); the integrated call set also achieved ≥95% concordance for 25 genes and improved or maintained concordance relative to the better‐performing single platform for 29 genes overall. The largest incremental gains among high‐concordance loci were observed for CYP1A1, CYP3A4, and DPYD, where integration improved concordance by 1.3–3.3 percentage points over the better‐performing single platform, consistent with complementary recovery of non-coding and rare haplotype‐defining variants. In contrast, UGT1A1 and CYP2D6 remained challenging despite integration. Discussion: For CYP2D6, the WGS analytical reference was generated by a single computational caller and was not orthogonally validated; concordance therefore reflects agreement with this reference rather than clinical typing accuracy. These findings provide practical gene‐level guidance for population‐scale pharmacogenetic analysis when WGS is unavailable.
- 26 October 2026
- Published by Frontiers Media SA in Frontiers in Pharmacology
- Vol. 17, 1892708
Open AccessIntegrating whole-exome and genotyping array data for pharmacogenetic typing: a comparative evaluation with whole-genome sequencing in UK Biobank
Introduction: Whole‐genome sequencing (WGS) provides comprehensive coverage of coding, non‐coding, and rare variants for pharmacogenetic star‐allele typing, but cost and computational demands limit its population-scale use. Whole-exome sequencing (WES) and imputed genotyping arrays are more widely available, yet their WGS-referenced performance remains incompletely defined. Methods: We benchmarked star-allele calls from WES, imputed array data, and an integrated call set combining WES and array data against an Aldy-derived WGS analytical reference across 33 pharmacogenes in 1,450 multi-ancestry UK Biobank participants. We also developed PGxIntegrator, an open‐source workflow that harmonizes WES and array data and enables scalable Aldy‐based star‐allele calling. Results: Using WES or imputed array data alone, 25 of 33 genes reached high diplotype concordance with the WGS analytical reference (≥95%); the integrated call set also achieved ≥95% concordance for 25 genes and improved or maintained concordance relative to the better‐performing single platform for 29 genes overall. The largest incremental gains among high‐concordance loci were observed for CYP1A1, CYP3A4, and DPYD, where integration improved concordance by 1.3–3.3 percentage points over the better‐performing single platform, consistent with complementary recovery of non-coding and rare haplotype‐defining variants. In contrast, UGT1A1 and CYP2D6 remained challenging despite integration. Discussion: For CYP2D6, the WGS analytical reference was generated by a single computational caller and was not orthogonally validated; concordance therefore reflects agreement with this reference rather than clinical typing accuracy. These findings provide practical gene‐level guidance for population‐scale pharmacogenetic analysis when WGS is unavailable.
- 29 September 2026
- Published by Oxford University Press (OUP) in American Journal of Epidemiology
Event-time regression discontinuity for biomedical data: local average treatment effect among the treated estimation and covariate distribution tests
Estimating causal effects of drugs on clinical biomarkers is often infeasible through randomized trials, while observational data require designs that approximate experimental conditions. Using linked UK Biobank records collected during 2006-2013, we develop an event-time regression discontinuity framework for settings in which prescription records are longitudinal but biomarkers are measured cross-sectionally. A surrogate running variable aligns individuals by the interval between biomarker measurement and the first recorded prescription, comparing eventual users measured shortly before vs after that date. Under continuity assumptions, this identifies a local average treatment effect among the treated. We apply distributional covariate-balance diagnostics, robust bias-corrected inference, and parsimonious covariate adjustment. Statin analyses show a local downward low-density lipoprotein discontinuity around the first recorded prescription date, with direction and statistical significance robust across bias-corrected inference, bandwidths, and donut specifications. However, the running-variable density test is significant and donut estimates vary from −1.00 to −1.72 mmol/L, indicating residual uncertainty about local comparability. The metformin–HbA1c example is less stable: a sizeable parametric local-linear reduction is not corroborated under robust bias-corrected inference. Event-time regression discontinuity offers potential for pharmacovigilance and real-world causal inference in linked health databases, but conclusions depend on explicit local timing-exogeneity and continuity assumptions.
- 29 September 2026
- Published by Oxford University Press (OUP) in American Journal of Epidemiology
Event-time regression discontinuity for biomedical data: local average treatment effect among the treated estimation and covariate distribution tests
Estimating causal effects of drugs on clinical biomarkers is often infeasible through randomized trials, while observational data require designs that approximate experimental conditions. Using linked UK Biobank records collected during 2006-2013, we develop an event-time regression discontinuity framework for settings in which prescription records are longitudinal but biomarkers are measured cross-sectionally. A surrogate running variable aligns individuals by the interval between biomarker measurement and the first recorded prescription, comparing eventual users measured shortly before vs after that date. Under continuity assumptions, this identifies a local average treatment effect among the treated. We apply distributional covariate-balance diagnostics, robust bias-corrected inference, and parsimonious covariate adjustment. Statin analyses show a local downward low-density lipoprotein discontinuity around the first recorded prescription date, with direction and statistical significance robust across bias-corrected inference, bandwidths, and donut specifications. However, the running-variable density test is significant and donut estimates vary from −1.00 to −1.72 mmol/L, indicating residual uncertainty about local comparability. The metformin–HbA1c example is less stable: a sizeable parametric local-linear reduction is not corroborated under robust bias-corrected inference. Event-time regression discontinuity offers potential for pharmacovigilance and real-world causal inference in linked health databases, but conclusions depend on explicit local timing-exogeneity and continuity assumptions.
- Open Access
- 8 September 2026
- Published by Springer Nature in Research Square
SPASQR applies smoothed quantile regression to detect heterogeneous genetic associations
Genome-wide association studies (GWAS) of quantitative traits are mostly based on linear regression, which associates genetic variants with the mean of the phenotype but has limited power to detect dynamic and context-dependent genetic effects. Quantile regression (QR) offers a powerful alternative to address this limitation, however, existing QR-based GWAS methods remain underdeveloped. Here, we introduce SPA SQR , a scalable and robust QR-based framework designed for biobank-scale GWAS. SPA SQR includes three key innovations: (i) convolution-smoothed quantile regression to enhance statistical power; (ii) score tests based on a retrospective framework to effectively control for relatedness; and (iii) saddlepoint approximation to accurately calibrate p-values for rare variants. In extensive simulations, SPA SQR outperformed existing QR methods in both type I error control and statistical power. Applied to 66 UK Biobank quantitative traits, SPA SQR detected 1,618 loci with heterogeneous effect patterns that were missed by conventional linear approaches. Notably, SPA SQR achieves over 100-fold speedup relative to existing QR-based methods, with computational efficiency comparable to popular GWAS tools such as REGENIE and LDAK-KVIK.
- 8 September 2026
- Published by Springer Nature in Research Square
Open AccessSPASQR applies smoothed quantile regression to detect heterogeneous genetic associations
Genome-wide association studies (GWAS) of quantitative traits are mostly based on linear regression, which associates genetic variants with the mean of the phenotype but has limited power to detect dynamic and context-dependent genetic effects. Quantile regression (QR) offers a powerful alternative to address this limitation, however, existing QR-based GWAS methods remain underdeveloped. Here, we introduce SPA SQR , a scalable and robust QR-based framework designed for biobank-scale GWAS. SPA SQR includes three key innovations: (i) convolution-smoothed quantile regression to enhance statistical power; (ii) score tests based on a retrospective framework to effectively control for relatedness; and (iii) saddlepoint approximation to accurately calibrate p-values for rare variants. In extensive simulations, SPA SQR outperformed existing QR methods in both type I error control and statistical power. Applied to 66 UK Biobank quantitative traits, SPA SQR detected 1,618 loci with heterogeneous effect patterns that were missed by conventional linear approaches. Notably, SPA SQR achieves over 100-fold speedup relative to existing QR-based methods, with computational efficiency comparable to popular GWAS tools such as REGENIE and LDAK-KVIK.
- Open Access
- 1 September 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (5)
DeepU-PRS: nonlinear and uncertainty-aware SNP heritability modeling for annotation-informed polygenic risk scoring
Polygenic risk scores (PRSs) are widely used to quantify genetic predisposition to complex traits, and their predictive accuracy can be enhanced by incorporating functional annotations into single-nucleotide polymorphism (SNP) effect-size estimation. Existing approaches typically assume linear relationships between annotations and SNP effects, neglecting predictive aleatoric variance and potential nonlinear regulatory interactions. To evaluate the gains of nonlinear and uncertainty-aware modeling, we developed DeepU-PRS, a deep ensemble framework that extends the linear BLD-LDAK model with a neural network trained on functional annotations to jointly estimate SNP heritability and its aleatoric uncertainty, which are then integrated into a Bayesian PRS model. Across 28 continuous and 8 disease phenotypes in the UK Biobank, DeepU-PRS improved the mean continuous R2 by approximately 10% over linear BLD-LDAK, with the largest gains for traits whose association signals were concentrated, whereas the more diffuse disease traits favored BLD-LDAK. These patterns were reproduced in individual-level data from All of Us. Against a panel of established methods (C + T, Lassosum, PRS-CS, LDpred-funct, and SBayesRC), DeepU-PRS was among the strongest single methods on continuous traits, and ensembles combining DeepU-PRS with the linear baseline and SBayesRC further improved accuracy, indicating complementary predictive signal. These findings highlight the value of modeling nonlinear and uncertain regulatory effects to advance PRS prediction.
- 1 September 2026
- Published by Oxford University Press (OUP) in Briefings in Bioinformatics
- Vol. 27 (5)
Open AccessDeepU-PRS: nonlinear and uncertainty-aware SNP heritability modeling for annotation-informed polygenic risk scoring
Polygenic risk scores (PRSs) are widely used to quantify genetic predisposition to complex traits, and their predictive accuracy can be enhanced by incorporating functional annotations into single-nucleotide polymorphism (SNP) effect-size estimation. Existing approaches typically assume linear relationships between annotations and SNP effects, neglecting predictive aleatoric variance and potential nonlinear regulatory interactions. To evaluate the gains of nonlinear and uncertainty-aware modeling, we developed DeepU-PRS, a deep ensemble framework that extends the linear BLD-LDAK model with a neural network trained on functional annotations to jointly estimate SNP heritability and its aleatoric uncertainty, which are then integrated into a Bayesian PRS model. Across 28 continuous and 8 disease phenotypes in the UK Biobank, DeepU-PRS improved the mean continuous R2 by approximately 10% over linear BLD-LDAK, with the largest gains for traits whose association signals were concentrated, whereas the more diffuse disease traits favored BLD-LDAK. These patterns were reproduced in individual-level data from All of Us. Against a panel of established methods (C + T, Lassosum, PRS-CS, LDpred-funct, and SBayesRC), DeepU-PRS was among the strongest single methods on continuous traits, and ensembles combining DeepU-PRS with the linear baseline and SBayesRC further improved accuracy, indicating complementary predictive signal. These findings highlight the value of modeling nonlinear and uncertain regulatory effects to advance PRS prediction.
- Open Access
- 17 August 2026
- Published by Springer Nature in Nature Genetics
- Vol. 58 (9) , 2381-2388
Rare variant effect estimation and polygenic risk prediction
- 17 August 2026
- Published by Springer Nature in Nature Genetics
- Vol. 58 (9) , 2381-2388
Open AccessRare variant effect estimation and polygenic risk prediction
- Open Access
- 4 June 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Integrative genomic analysis of 21 orofacial diseases identifies shared genetic architecture with systemic diseases
Orofacial diseases are closely linked to systemic health, yet their genetic architecture and causal relationships remain incompletely understood. Here we show that integrative genomic analyses of 21 orofacial diseases in the UK Biobank, combining genome-wide, transcriptome-wide, rare variant association studies and Mendelian randomization, identify 48 novel loci and prioritize 160 putative causal genes. Among these, a synonymous variant in ALDH1A2 associated with periodontitis, encoding a key enzyme in retinoic acid synthesis, reduces translational efficiency, as shown by CRISPR-Cas9 editing. Single-cell RNA sequencing of periodontitis tissues further reveals impaired retinoic acid signaling and a T helper 17-skewed immune phenotype, accompanied by reduced retinoic acid levels in gingival crevicular fluid. Mendelian randomization supports a causal effect of Sjögren's syndrome on lung cancer and stroke, replicated in FinnGen. By contrast, dental caries and periodontitis share substantial genetic architecture with systemic diseases, including enrichment of nicotine-related pathways, without robust evidence of causality. Together, these findings map the genetic landscape of orofacial diseases and clarify their complex links to systemic pathophysiology.
- 4 June 2026
- Published by Springer Nature in Nature Communications
- Vol. 17 (1)
Open AccessIntegrative genomic analysis of 21 orofacial diseases identifies shared genetic architecture with systemic diseases
Orofacial diseases are closely linked to systemic health, yet their genetic architecture and causal relationships remain incompletely understood. Here we show that integrative genomic analyses of 21 orofacial diseases in the UK Biobank, combining genome-wide, transcriptome-wide, rare variant association studies and Mendelian randomization, identify 48 novel loci and prioritize 160 putative causal genes. Among these, a synonymous variant in ALDH1A2 associated with periodontitis, encoding a key enzyme in retinoic acid synthesis, reduces translational efficiency, as shown by CRISPR-Cas9 editing. Single-cell RNA sequencing of periodontitis tissues further reveals impaired retinoic acid signaling and a T helper 17-skewed immune phenotype, accompanied by reduced retinoic acid levels in gingival crevicular fluid. Mendelian randomization supports a causal effect of Sjögren's syndrome on lung cancer and stroke, replicated in FinnGen. By contrast, dental caries and periodontitis share substantial genetic architecture with systemic diseases, including enrichment of nicotine-related pathways, without robust evidence of causality. Together, these findings map the genetic landscape of orofacial diseases and clarify their complex links to systemic pathophysiology.
- Open Access
- 3 April 2026
- Published by Springer Nature in Research Square
Discovery of gene-alcohol interaction loci influencing blood pressure in 1.1 million individuals from multiple populations
Genetic predisposition and alcohol consumption are risk factors for increased blood pressure (BP), but their interactions influencing BP remain understudied. We conducted population-specific and cross-population meta-analyses of genome-wide gene-alcohol (GxAlc) interactions affecting BP in >1.1M individuals from multiple populations. We identified 46 GxAlc interaction loci for BP, including 21 from one-degree-of-freedom interaction tests (PGxAlc<5x10-8; or <0.05/Meff, Meff independent BP associations at P<10-5), and 25 from two-degree-of-freedom tests of main and interaction effects (PGxAlc<0.05/M2df, M2df independent 2df-associations at P2df<5x10-8), including 7 novel and 39 known BP loci. The 12q24 locus highlights the genetic effect of BRAP-rs11066001 on BP, being ~6 times larger in current drinkers than in non-drinkers. Gene prioritization with 46 GxAlc loci identified 15 genes with ≥3 lines of evidence (location, literature, druggability, functional/regulatory annotation, or pathway analyses). Several loci showed sex- and population-specific effects and revealed biological pathways of alcohol’s influence on BP, suggesting mechanisms underlying alcohol-induced hypertension.
- 3 April 2026
- Published by Springer Nature in Research Square
Open AccessDiscovery of gene-alcohol interaction loci influencing blood pressure in 1.1 million individuals from multiple populations
Genetic predisposition and alcohol consumption are risk factors for increased blood pressure (BP), but their interactions influencing BP remain understudied. We conducted population-specific and cross-population meta-analyses of genome-wide gene-alcohol (GxAlc) interactions affecting BP in >1.1M individuals from multiple populations. We identified 46 GxAlc interaction loci for BP, including 21 from one-degree-of-freedom interaction tests (PGxAlc<5x10-8; or <0.05/Meff, Meff independent BP associations at P<10-5), and 25 from two-degree-of-freedom tests of main and interaction effects (PGxAlc<0.05/M2df, M2df independent 2df-associations at P2df<5x10-8), including 7 novel and 39 known BP loci. The 12q24 locus highlights the genetic effect of BRAP-rs11066001 on BP, being ~6 times larger in current drinkers than in non-drinkers. Gene prioritization with 46 GxAlc loci identified 15 genes with ≥3 lines of evidence (location, literature, druggability, functional/regulatory annotation, or pathway analyses). Several loci showed sex- and population-specific effects and revealed biological pathways of alcohol’s influence on BP, suggesting mechanisms underlying alcohol-induced hypertension.
- Open Access
- 20 November 2025
- Published by Springer Nature in Nature Genetics
- Vol. 57 (12) , 3185-3192
Scalable and accurate rare variant meta-analysis with Meta-SAIGE
Meta-analysis enhances the power of rare variant association tests by combining summary statistics across several cohorts. However, existing methods often fail to control type I error for low-prevalence binary traits and are computationally intensive. Here we introduce Meta-SAIGE—a scalable method for rare variant meta-analysis that accurately estimates the null distribution to control type I error and reuses the linkage disequilibrium matrix across phenotypes to boost computational efficiency in phenome-wide analyses. Simulations using UK Biobank whole-exome sequencing data show that Meta-SAIGE effectively controls type I error and achieves power comparable to pooled individual-level analysis with SAIGE-GENE+. Applying Meta-SAIGE to 83 low-prevalence phenotypes in UK Biobank and All of Us whole-exome sequencing data identified 237 gene–trait associations. Notably, 80 of these associations were not significant in either dataset alone, underscoring the power of our meta-analysis.
- 20 November 2025
- Published by Springer Nature in Nature Genetics
- Vol. 57 (12) , 3185-3192
Open AccessScalable and accurate rare variant meta-analysis with Meta-SAIGE
Meta-analysis enhances the power of rare variant association tests by combining summary statistics across several cohorts. However, existing methods often fail to control type I error for low-prevalence binary traits and are computationally intensive. Here we introduce Meta-SAIGE—a scalable method for rare variant meta-analysis that accurately estimates the null distribution to control type I error and reuses the linkage disequilibrium matrix across phenotypes to boost computational efficiency in phenome-wide analyses. Simulations using UK Biobank whole-exome sequencing data show that Meta-SAIGE effectively controls type I error and achieves power comparable to pooled individual-level analysis with SAIGE-GENE+. Applying Meta-SAIGE to 83 low-prevalence phenotypes in UK Biobank and All of Us whole-exome sequencing data identified 237 gene–trait associations. Notably, 80 of these associations were not significant in either dataset alone, underscoring the power of our meta-analysis.
- Open Access
Preference-Based Fine-Tuning of Genomic Sequence Models for Personal Expression Prediction with Data Augmentation
Despite substantial progress in genomic foundation models, accurately predicting inter-individual variation in gene expression from DNA sequence alone remains a major challenge. Current sequence-based models, such as Enformer and Borzoi, trained exclusively on the reference genome, cannot capture the effects of individual-specific regulatory variants. Moreover, the acquisition of paired whole-genome and transcriptome data required for personalized modeling is hindered by privacy and data-sharing constraints. To address this limitation, we integrate genomic data synthesis with established statistical frameworks. Our approach generates thousands of synthetic training samples by simulating genetic variation from the 1000 Genomes Project and assigning pseudo-expression labels using PrediXcan, a validated eQTL-based predictor. Because simulated and real expression values differ in scale and distribution, we introduce a preference-based objective that models relative rather than absolute expression patterns. Fine-tuning Enformer through alternating cycles of real-data regression and synthetic-data preference optimization enables efficient learning from both real and synthesized data. Using the GEUVADIS dataset, our framework outperforms AlphaGenome, PrediXcan, and Enformer fine-tuned without synthesized data, demonstrating that simulation-based integration of population-level regulatory knowledge can effectively mitigate data scarcity and improve cross-individual generalization in sequence-based gene expression prediction. Availability and implementation: Code and data are available at https://github.com/pacifiic/augment-finetune-genomics .
Open AccessPreference-Based Fine-Tuning of Genomic Sequence Models for Personal Expression Prediction with Data Augmentation
Despite substantial progress in genomic foundation models, accurately predicting inter-individual variation in gene expression from DNA sequence alone remains a major challenge. Current sequence-based models, such as Enformer and Borzoi, trained exclusively on the reference genome, cannot capture the effects of individual-specific regulatory variants. Moreover, the acquisition of paired whole-genome and transcriptome data required for personalized modeling is hindered by privacy and data-sharing constraints. To address this limitation, we integrate genomic data synthesis with established statistical frameworks. Our approach generates thousands of synthetic training samples by simulating genetic variation from the 1000 Genomes Project and assigning pseudo-expression labels using PrediXcan, a validated eQTL-based predictor. Because simulated and real expression values differ in scale and distribution, we introduce a preference-based objective that models relative rather than absolute expression patterns. Fine-tuning Enformer through alternating cycles of real-data regression and synthetic-data preference optimization enables efficient learning from both real and synthesized data. Using the GEUVADIS dataset, our framework outperforms AlphaGenome, PrediXcan, and Enformer fine-tuned without synthesized data, demonstrating that simulation-based integration of population-level regulatory knowledge can effectively mitigate data scarcity and improve cross-individual generalization in sequence-based gene expression prediction. Availability and implementation: Code and data are available at https://github.com/pacifiic/augment-finetune-genomics .
- Open Access
- 24 September 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (10)
PTL-PRS: an R package for transfer learning of polygenic risk scores with pseudovalidation
Summary: Polygenic risk scores (PRSs) are essential tools for predicting individual phenotypic risk but often lack accuracy in non-European ancestry groups. Transfer Learning for Polygenic Risk Scores (TL-PRS) addresses this challenge by leveraging European PRSs to improve prediction in underrepresented ancestries but requires privacy-sensitive individual-level data and has low computational efficiency. Therefore, we introduce Pseudovalidated Transfer Learning for PRS (PTL-PRS), an extension of TL-PRS that incorporates pseudovalidation to eliminate the need for individual-level data and includes further software optimization. For pseudovalidation, PTL-PRS generates pseudo-summary statistics for training and validation and evaluates model performance with the pseudo-R2 metric. To improve computational efficiency, PTL-PRS software was optimized with C++, blockwise early stopping, and direct genotype retrieval. Overall, PTL-PRS enhances usability while maintaining TL-PRS’s predictive performance. Availability and implementation: The PTL.PRS R package is publicly available on GitHub at https://github.com/bokeumcho/PTL.PRS. The summary statistics used in this paper are available in the public domain: UK Biobank (https://pheweb.org/UKB-TOPMed), PGS Catalog (https://www.pgscatalog.org), COVID-19 Host Genetics Initiative (https://www.covid19hg.org) and GenOMICC (https://genomicc.org/data).
- 24 September 2025
- Published by Oxford University Press (OUP) in Bioinformatics
- Vol. 41 (10)
Open AccessPTL-PRS: an R package for transfer learning of polygenic risk scores with pseudovalidation
Summary: Polygenic risk scores (PRSs) are essential tools for predicting individual phenotypic risk but often lack accuracy in non-European ancestry groups. Transfer Learning for Polygenic Risk Scores (TL-PRS) addresses this challenge by leveraging European PRSs to improve prediction in underrepresented ancestries but requires privacy-sensitive individual-level data and has low computational efficiency. Therefore, we introduce Pseudovalidated Transfer Learning for PRS (PTL-PRS), an extension of TL-PRS that incorporates pseudovalidation to eliminate the need for individual-level data and includes further software optimization. For pseudovalidation, PTL-PRS generates pseudo-summary statistics for training and validation and evaluates model performance with the pseudo-R2 metric. To improve computational efficiency, PTL-PRS software was optimized with C++, blockwise early stopping, and direct genotype retrieval. Overall, PTL-PRS enhances usability while maintaining TL-PRS’s predictive performance. Availability and implementation: The PTL.PRS R package is publicly available on GitHub at https://github.com/bokeumcho/PTL.PRS. The summary statistics used in this paper are available in the public domain: UK Biobank (https://pheweb.org/UKB-TOPMed), PGS Catalog (https://www.pgscatalog.org), COVID-19 Host Genetics Initiative (https://www.covid19hg.org) and GenOMICC (https://genomicc.org/data).