PGS Catalog - PGS005388 / Paroxysmal atrial fibrillation (Polygenic Score)

Polygenic Score (PGS) ID: PGS005388

Predicted Trait
Reported Trait	Paroxysmal atrial fibrillation
Mapped Trait(s)	Paroxysmal atrial fibrillation (HP_0004757)

Released in PGS Catalog: June 17, 2026

Download Score FTP directory

Terms and Licenses

PGS obtained from the Catalog should be cited appropriately, and used in accordance with any licensing restrictions set by the authors. See EBI Terms of Use (https://www.ebi.ac.uk/about/terms-of-use/) for additional details.

Score Details

Score Construction
PGS Name	PAF_PRS
Development Method
Name	LD-clumping and p-value thresholding (P+T) using GWAS summary statistics
Parameters	Variants for the score were selected from a large-scale Japanese atrial fibrillation GWAS (Low et al., Nat Genet 2017, doi:10.1038/ng.3842). Using these summary statistics, we constructed 30 candidate SNP sets by combining three P-value thresholds (5×10-8, 1×10-5, 1×10-4) with ten LD clumping thresholds (r2 = 0.9-0.1 and no clumping) in PLINK 1.9. The final PAF_PRS used 122 SNPs selected at P < 1×10-5 and r2 > 0.3, which achieved the highest mean cross-validated AUC in the 90% training dataset.
Variants
Original Genome Build	GRCh37
Number of Variants	122
Effect Weight Type	beta

PGS Source
PGS Catalog Publication (PGP) ID	PGP000775
Citation (link to publication)	Shiomi M et al. PLoS One (2026)

Ancestry Distribution
Source of Variant Associations (GWAS)	East Asian: 100% 36,792 individuals (100%)
Score Development/Training	East Asian: 100% 1,603 individuals (100%)
PGS Evaluation	East Asian: 100% 1 Sample Sets

Development Samples

Source of Variant Associations (GWAS)

Study Identifiers	Sample Numbers	Sample Ancestry	Cohort(s)
GWAS Catalog: GCST004373 Europe PMC: 28416822	36,792 individuals	East Asian	NR

Score Development/Training

Study Identifiers	Sample Numbers	Sample Ancestry	Cohort(s)	Phenotype Definitions & Methods	Age of Study Participants	Participant Follow-up Time	Additional Ancestry Description	Additional Sample/Cohort Information
—	1,603 individuals [ 934 cases , 669 controls ] , 63.0 % Male samples	East Asian (Japanese)	PAF_JPN	Paroxysmal atrial fibrillation (PAF) cases and controls recruited from cardiology departments. Exclusion criteria were: (1) missing clinical information, (2) AF other than PAF, (3) history of heart failure, (4) coronary artery disease, or (5) valvular heart disease. Phenotypes based on clinical records and ECG findings.	—	—	—	Training dataset consisting of 90% of PAF cases and controls, obtained by stratified sampling to preserve the case-control ratio. preserve the case-control ratio.

Performance Metrics

Disclaimer: The performance metrics are displayed as reported by the source studies. It is important to note that metrics are not necessarily comparable with each other. For example, metrics depend on the sample characteristics (described by the PGS Catalog Sample Set [PSS] ID), phenotyping, and statistical modelling. Please refer to the source publication for additional guidance on performance.

PGS Performance Metric ID (PPM)	PGS Sample Set ID (PSS)	Performance Source	Trait	PGS Effect Sizes (per SD change)	Classification Metrics	Other Metrics	Covariates Included in the Model	PGS Performance: Other Relevant Information
PPM023426	PSS012177\| East Asian Ancestry\| 179 individuals	PGP000775 \| Shiomi M et al. PLoS One (2026)	Reported Trait: Paroxysmal atrial fibrillation	—	AUROC: 0.702 [0.624, 0.77]	AUPRC: 0.776 F1: 0.698	—	Best PRS-only model (Light Gradient Boosting Machine, LGBM) using the 122-SNP PAF_PRS at P < 1 × 10-5 and r2 > 0.3; performance evaluated on the 10% test dataset (PAF_JPN_test) with 1,000 bootstrap resamples.
PPM023427	PSS012177\| East Asian Ancestry\| 179 individuals	PGP000775 \| Shiomi M et al. PLoS One (2026)	Reported Trait: Paroxysmal atrial fibrillation	—	AUROC: 0.737 [0.668, 0.804]	AUPRC: 0.781 F1: 0.748	sex, smoking, alcohol consumption, history of hypertension, diabetes mellitus, dyslipidemia, cerebral infarction, and family history of AF	Best additive model (Extreme Gradient Boosting, XGB) with the 122-SNP PAF_PRS and the listed clinical covariates; same P-value and LD r2 thresholds and test dataset as the PRS-only model; performance metrics obtained from 1,000 bootstrap resamples.

Evaluated Samples

PGS Sample Set ID (PSS)	Phenotype Definitions and Methods	Participant Follow-up Time	Sample Numbers	Age of Study Participants	Sample Ancestry	Additional Ancestry Description	Cohort(s)	Additional Sample/Cohort Information
PSS012177	Paroxysmal atrial fibrillation (PAF) cases and controls recruited from cardiology departments. Exclusion criteria were: (1) missing clinical information, (2) AF other than PAF, (3) history of heart failure, (4) coronary artery disease, or (5) valvular heart disease. Phenotypes based on clinical records and ECG findings.	—	179 individuals [ 104 cases , 75 controls ] , 65.0 % Male samples	—	East Asian (Japanese)	—	PAF_JPN	Held-out 10% test dataset obtained by stratified sampling to preserve the case-control ratio.