polygenic scores

This statistical review comprehensively analyzes 46 methods for Polygenic Score (PGS) construction, unifying most of them under a multiple linear regression framework to clarify their assumptions regarding effect size distribution and Linkage Disequilibrium (LD).
The review concludes that optimal PGS performance (accuracy) is highly dependent on the trait’s genetic architecture and is significantly improved by using Bayesian/Regularization methods (e.g., LDpred, PRS-CS) that explicitly model LD and incorporate informed prior distributions for SNP effects.
A critical challenge highlighted is the significant loss of transferability across ancestral populations, underscoring the need for more diverse training data and methods that better address ancestral heterogeneity and incorporate non-additive and Gene-by-Environment (GxE) effects.

Goal: To introduce LDpred-funct, a polygenic prediction method that uses trait-specific functional priors to enhance prediction accuracy.
Method: LDpred-funct incorporates information from the baseline-LD model (including coding, regulatory, and conserved annotations) to estimate posterior mean causal effect sizes, followed by a cross-validation regularization step to account for genetic sparsity.
Finding: The method achieved a +4.6% relative improvement in average prediction \(R^2\) across 21 UK Biobank traits compared to the best non-functional method. This demonstrates that leveraging the functional architecture of the genome leads to more accurate Polygenic Risk Scores (PRS).

This large-scale study evaluated the cross-ancestry transferability of Polygenic Risk Scores (PRSs) for four common diseases (CAD, T2D, breast, and prostate cancer) using data from six biobanks and over one million individuals of diverse global ancestries.
The analysis found that PRS transferability was high and robust across different populations and substructures of European ancestry, but was significantly lower for individuals of African, South Asian, and East Asian ancestry.
The poor transferability, which was most pronounced in African ancestry individuals, highlights the critical issue of ancestral bias in genomic research and the potential for current PRS implementation to exacerbate health disparities until more diverse training data are available.

Core Principle: This study partitioned the genetic heterogeneity of Type 2 Diabetes (T2D) using a novel colocalization-first approach followed by network-based clustering of T2D and 20 related metabolic traits across 243 loci.
Key Finding: The method identified five distinct T2D biological pathways (Obesity, Lipodystrophic insulin resistance, Liver/lipid metabolism, Hepatic glucose metabolism, and Beta-cell dysfunction), successfully isolating genetically distinct disease mechanisms.
Clinical Significance: Partitioned Polygenic Risk Scores (PRSs) showed heterogeneous clinical associations in a validation cohort (n=21,742 T2D individuals); notably, the Lipodystrophic insulin resistance PRS and Beta-cell dysfunction PRS were causally associated with lower BMI, providing genetic validation for the clinically important “lean diabetes” sub-type.
Methodological Advance: By integrating colocalization and Mendelian Randomization, the framework provided stronger inferences on the causality and directionality of the genetic associations, which is essential for translating genetic discoveries into targeted T2D treatments.

Core Principle: This landmark study challenged the traditional use of discrete ancestry groups for polygenic scores (PGSs), proposing a framework that evaluates individual-level PGS accuracy based on Genetic Distance (GD) from the GWAS training population.
Key Finding: They demonstrated that individual-level PGS accuracy experiences a continuous, steep decay as GD from the training data increases, with an average Pearson correlation of R = -0.95 across 84 complex traits, confirming that performance loss is a predictable function of genetic dissimilarity.
Health Equity Implication: The study quantified a major health disparity, showing that the genetically closest individuals of non-European ancestry (e.g., Hispanic/Latino American) had PGS accuracy comparable to the most distant European-ancestry individuals, highlighting the severe and systematic bias due to lack of diversity in training cohorts.
Recommendation: The authors advocate for moving beyond discrete ancestry labels and using continuous metrics like GD to characterize and correct for performance disparities, thus ensuring more equitable clinical translation of PGSs.