ePRS

Summary

ePRS (external Polygenic Risk Score) is a transfer learning penalization framework designed to enhance the clinical utility of polygenic scores (PRSs) in small, phenotypically refined clinical target cohorts. By integrating individual-level target data with summary statistics from large-scale, heterogeneous external source GWAS, ePRS improves predictive performance and model stability. It employs a weighted elastic net formulation that incorporates a continuous measure of statistical evidence and genetic correlation as a fail-safe mechanism.

Method and Capabilities

ePRS addresses the power-specificity trade-off in polygenic risk score development by combining the strengths of large biobank-scale studies (which have large samples but dilute phenotypic specificity) and small clinical cohorts (which are deeply phenotyped but underpowered). Key capabilities and methodological components include: - Continuous Penalty Function: Rather than applying arbitrary P-value thresholds, ePRS uses a continuous measure of statistical evidence derived from external GWAS (typically $-\log_{10}p_j$) to define variant-specific penalty weights. Detailed formulations are documented in the ePRS Penalization Framework. - Genetic Correlation Integration: The framework scales the external evidence using the genetic correlation ($r_g$) between the source and target phenotypes. If $r_g \to 0$, ePRS downweights the external source phenotype, acting as a fail-safe that reduces to standard elastic net regression on the target cohort. - Covariate Adjustment: Covariates such as age, sex, and principal components are modeled as unpenalized fixed effects, ensuring robust control for confounders. - Multivariate Modeling: The method fits a penalized regression model directly to target cohort genotype matrices using iterative gradient descent.

Applications and Validation

ePRS has been validated across three distinct real-world clinical applications: 1. Epilepsy Subtype Differentiation: Refined a broad Idiopathic Generalized Epilepsy (IGE) PRS using data from the BIOJUME Consortium to improve prediction and differentiation of Juvenile Myoclonic Epilepsy (JME) from other subtypes. 2. Impulsivity in JME: Leveraged a large Attention Deficit Hyperactivity Disorder (ADHD) GWAS as an external source phenotype to enhance prediction of trait impulsivity (measured via the Barratt Impulsivity Scale) in JME patients. 3. Cystic Fibrosis-Related Diabetes (CFRD): Borrowed Type 2 Diabetes (T2D) summary statistics from the UK Biobank to predict CFRD onset within the Canadian Cystic Fibrosis Gene Modifier Study (CGMS) cohort, demonstrating robustness against cross-cohort genotyping and quality control artifacts.

The source code for ePRS is implemented in R and is available on GitHub (as of 2026-07-19).

Citations

• Lin, Y., Beier, C. P., Sobiskova, Z., Hamandi, K., Stödberg, T., Ng, C. C., Andrade, D. M., Syvertsen, M. R., Gardella, E., Orsini, A., Fong, C. Y., Zarubova, J., Kajsová, M., Lim, K. S., Selmer, K. K., Cerulli Irelli, E., Rubboli, G., Striano, P., BIOJUME Consortium, Pal, D. K., & Strug, L. J. (2026). Transfer learning enhances clinical utility of polygenic scores with small, phenotypically refined cohorts. Genome Research, 36(7), 1403-1416. Source paper: Genome Res.-2026-Lin-1403-16.pdf