LDpred2

LDpred2 computes polygenic scores from GWAS summary statistics and linkage disequilibrium (LD) correlation matrices to predict complex traits from genetic variants.

Key Features:

• Enhanced Predictive Performance: LDpred2 increases robustness and accuracy of polygenic score prediction relative to LDpred by refining how genetic correlations are utilized.
• Sparse Option: The "sparse" option identifies variant effects that are exactly zero to distinguish non-impacting from impacting variants.
• Auto Option: The "auto" option automatically learns two LDpred parameters directly from the data.
• Benchmarking Success: In simulated and real datasets LDpred2 outperformed LDpred1, lassosum, PRS-CS, and SBayesR, achieving a mean AUC of 65.1% across eight real traits.
• Genome-Wide Application: LDpred2 is recommended for genome-wide application rather than per-chromosome runs.

Scientific Applications:

• Complex trait prediction: Compute polygenic scores to predict complex traits and diseases from genetic variants.
• Heritability and genetic architecture: Support investigation of trait heritability and the genetic architecture underlying complex traits.
• Personalized medicine: Provide more precise genetic risk predictions to inform personalized medicine approaches.
• Implementation: LDpred2 is implemented within the R package bigsnpr.

Methodology:

LDpred2 derives polygenic scores from GWAS summary statistics and a matrix of correlations between genetic variants (LD matrix); the "sparse" option enforces exact-zero effect sizes for some variants and the "auto" option learns two LDpred parameters from the data.