Polygenic Risk Score (PRS)

The foundational and most computationally efficient PRS method. It first clumps variants using linkage disequilibrium (LD) to select independent index SNPs, then applies a p-value threshold to retain only variants meeting a significance cutoff.

• Clumping: Retains the most significant SNP within an LD window, pruning correlated neighbors
• Thresholding: Tests multiple p-value cutoffs (e.g., 5e-8, 1e-5, 0.1) to find the optimal score
• Scoring: Calculates the sum of risk alleles weighted by their GWAS effect sizes
• Limitation: The hard threshold discards many truly associated variants with small effects, reducing predictive power for highly polygenic traits

A regularized linear regression method that performs simultaneous variant selection and effect estimation. The L1 penalty forces many coefficient estimates to exactly zero, creating a sparse PRS model.

• Tuning parameter λ controls the strength of penalization and the number of retained variants
• Selected through cross-validation to maximize out-of-sample prediction
• Naturally handles LD by selecting one representative variant from correlated groups
• Computationally intensive with individual-level data but produces highly interpretable sparse models
• Elastic net extends LASSO by adding an L2 ridge penalty, improving stability when variants are highly correlated

A statistical framework that estimates the prior distribution of genetic effects directly from the observed GWAS data, enabling adaptive shrinkage without specifying a rigid parametric prior.

• SBayesR assumes a mixture of normal distributions with different variances, including a spike at zero
• MegaPRS models effect sizes as a function of LD, minor allele frequency, and functional annotations
• Estimates the genetic architecture parameters (polygenicity, effect size distribution) from summary statistics
• Provides posterior inclusion probabilities for each variant, useful for fine-mapping alongside prediction
• Particularly robust when the true genetic architecture deviates from standard modeling assumptions

Specialized frameworks addressing the portability gap where PRS trained in European populations lose substantial accuracy in non-European cohorts due to differences in LD structure and allele frequencies.

• PRS-CSx: Integrates GWAS summary statistics from multiple populations with shared continuous shrinkage priors
• XPASS: Models population-specific effect sizes as correlated random effects across ancestries
• PolyPred: Combines cross-population effect estimates with functional genomic annotations
• GAUDI: Uses penalized regression to fuse multiple ancestry-specific scores into an optimized composite
• Critical for equitable deployment of PRS in diverse clinical settings and global biobank studies

The foundational method for constructing a PRS. It selects independent genetic variants by pruning based on linkage disequilibrium and retaining only those below a specified p-value significance threshold.

• Clumping: Groups correlated variants into loci and selects the most significant variant per locus.
• Thresholding: Filters variants based on a p-value cutoff (e.g., p < 5e-8, p < 0.01).
• Scoring: The PRS is the sum of the risk allele count multiplied by the GWAS effect size for retained variants.
• Computationally efficient and requires only GWAS summary statistics and a reference LD panel.

A Bayesian PRS method that models the genetic architecture using a point-normal mixture prior on variant effect sizes. It does not rely on explicit p-value thresholding.

• Assumes a fraction of variants have non-zero effects, while the rest have zero effect.
• Uses a Gibbs sampler to infer posterior mean effect sizes from GWAS summary statistics and an external LD reference panel.
• The 'auto' version estimates the heritability and the fraction of causal variants directly from the data.
• Generally outperforms C+T when the genetic architecture is highly polygenic.

A Bayesian polygenic prediction method that applies continuous shrinkage priors on SNP effect sizes. It places a global-local scale mixture of normals prior to flexibly model varying genetic architectures.

• Uses GWAS summary statistics and an external LD reference panel to infer posterior effect sizes.
• The phi parameter controls the global sparsity, automatically adapting to the genetic architecture of the trait.
• Computationally efficient compared to MCMC-based methods like LDpred2.
• Well-suited for highly polygenic traits where many variants have small, non-zero effects.

A statistical framework for modeling binary disease traits. It assumes a continuous, unobserved liability underlying the observed case/control status.

• Individuals exceeding a fixed threshold on the liability scale are classified as affected cases.
• Allows heritability and variance explained (R²) to be estimated on the continuous liability scale, independent of disease prevalence.
• Essential for converting PRS R² on the observed scale to the liability scale for fair comparisons across traits with different prevalences.
• Forms the theoretical basis for many PRS evaluation metrics in case-control studies.

A polygenic risk score developed and validated across diverse global populations to address the critical issue of portability.

• Standard PRS trained on European GWAS often show substantial attenuation in African, Asian, and Hispanic/Latino populations.
• Methods include multi-ancestry meta-analysis GWAS, joint modeling of LD differences, and ancestry-specific tuning of effect sizes.
• PRS-CSx extends PRS-CS to jointly model multiple ancestry groups by coupling effect size distributions.
• Essential for equitable clinical implementation and avoiding health disparities in genomic medicine.

A metric quantifying the clinical utility of adding a PRS to an existing risk model. It measures the extent to which the new model correctly reassigns individuals to more appropriate risk categories.

• Event NRI: Focuses on cases, measuring the proportion correctly moved to a higher risk category minus those incorrectly moved lower.
• Non-event NRI: Focuses on controls, measuring the proportion correctly moved to a lower risk category.
• Category-free NRI: Evaluates any upward or downward change in predicted risk without predefined thresholds.
• Goes beyond discrimination (AUC) to assess whether a PRS meaningfully changes clinical decision-making.