PRSet

Summary

PRSet is a pathway-based polygenic risk scoring method, implemented as a module of PRSice, that computes a separate polygenic risk score (PRS) for each of thousands of predefined gene sets/pathways instead of aggregating all genome-wide variants into a single score. It also outputs a competitive permutation-based P value testing whether a pathway's PRS predicts a trait better than randomly matched gene sets, giving individual-level pathway liability estimates alongside enrichment testing.

Method

• For pathway k, PRSet computes PRS_ik = Σβ_j·G_ij over the clumped SNPs mapped to that pathway's genes (β_j from GWAS summary statistics, G_ij the individual's genotype dosage), using independent clumping within each pathway rather than genome-wide clumping — this retains pathway-specific signal and accounts for correlation between SNPs in nearby genes.[1]
• Competitive significance for each pathway PRS is assessed via permutation (10,000 permutations in the paper's benchmarks) against randomly sampled gene sets of matched size, rather than relying on the self-contained P value alone.[1]
• The paper's pathway catalogue drew on 4,079 pathways/gene sets across six databases (KEGG, BioCarta, Reactome, the Pathway Interaction Database, Gene Ontology, and the Mouse Genome Database), filtered to gene sets of 10–2,000 genes, with gene coordinates extended 35 kb upstream / 10 kb downstream to capture regulatory regions.[1]

Benchmarks

• Enrichment concordance: on six real traits (LDL cholesterol, coronary artery disease, schizophrenia, BMI, Alzheimer's disease, alcohol consumption), PRSet's pathway enrichment broadly agreed with MAGMA and LD-score regression (modest Kendall's τ against expert-curated MalaCards relevance scores), but the methods diverged on specific cases: MAGMA/LDSC correlated better with expert opinion for schizophrenia tissue enrichment, while PRSet alone flagged microglia-specific enrichment for Alzheimer's disease under its top-quantile test — the only method to do so.[1]
• Disease stratification: across 21 test scenarios — 2 real clinical/molecular subtype pairs (Crohn's disease vs. ulcerative colitis, N=5,782 UK Biobank; bipolar I vs. II, N=14,459 across 21 Psychiatric Genomics Consortium cohorts), 10 "pseudo-subtype" pairings built from 5 UK Biobank diseases (T2D, CAD, hypercholesterolemia, obesity, hypertension; N=19,668–92,818 per disease), and 9 comorbidity-presence/absence scenarios — PRSet outperformed standard genome-wide PRS (from clumping+thresholding and lassosum) in 20 of 21 scenarios.[1] Reported discriminative R² was modest in absolute terms (e.g. R² = 9.27×10⁻³ for CD vs. UC, R² = 0.032 for bipolar I vs. II, R² < 0.016 for comorbidity scenarios), but consistently favored PRSet over the genome-wide comparators.[1]

See Also

• Mechanism-First Polygenic Risk Score Clustering — the broader family of pathway/mechanism-based approaches to disease subtyping PRSet belongs to
• Polygenic Risk Scores
• Proteogenomic Classification of IBD Subtypes — an independent CD-vs-UC classification study using a standard genome-wide PRS combined with proteomics, rather than PRSet's pathway scores

Citations

[1] Choi, S. W., García-González, J., Ruan, Y., Wu, H. M., Porras, C., Johnson, J., Bipolar Disorder Working Group of the Psychiatric Genomics Consortium, Hoggart, C. J., & O'Reilly, P. F. (2023). PRSet: Pathway-based polygenic risk score analyses and software. PLOS Genetics, 19(2), e1010624. DOI: 10.1371/journal.pgen.1010624. Supports: method description, pathway catalogue, and all benchmark/stratification results above. Location: Full text — Methods and Results sections.