VIPRS

VIPRS applies variational Bayesian inference to estimate joint single-nucleotide polymorphism (SNP) effect sizes from GWAS summary statistics and SNP array data for polygenic risk score (PRS) modeling and phenotype prediction.

Key Features:

• Bayesian Framework: Employs a Bayesian model to infer joint effect sizes of genetic variants using GWAS summary statistics.
• Variational Inference: Uses variational inference to approximate posterior distributions of effect sizes as an alternative to Markov chain Monte Carlo (MCMC), reducing computational time and resource requirements.
• Performance Efficiency: Demonstrated competitive prediction accuracy across 36 simulation configurations and 12 UK Biobank phenotypes while running over twofold faster than MCMC-based approaches.
• Scalability: Scales to large marker sets, validated on datasets comprising 9.6 million genetic markers and applied to highly polygenic traits such as height.
• Transferability Across Ethnic Groups: Improves cross-population PRS transferability, achieving up to a 1.7-fold increase in R² for LDL cholesterol prediction in individuals of Nigerian ancestry relative to White British samples.

Scientific Applications:

• Genetic Epidemiology: Enables inference of genetic architecture and PRS construction for complex traits using large-scale GWAS summary statistics and SNP array data.
• Personalized Medicine: Supports phenotype prediction and risk assessment for complex diseases via PRS-based models.
• Cross-population PRS Evaluation: Facilitates assessment and improvement of PRS transferability across different ancestral groups, demonstrated with UK Biobank and Nigerian ancestry samples.

Methodology:

Infers joint SNP effect sizes under a Bayesian model using GWAS summary statistics and SNP array data, approximates posterior distributions with variational inference instead of MCMC, and benchmarks performance across 36 simulation settings and 12 UK Biobank phenotypes including tests on 9.6 million markers.