Sparse polygenic risk score inference with the spike-and-slab LASSO

Area of Science:

• Genetics
• Bioinformatics
• Statistical genetics

Background:

• Large-scale biobanks offer rich data for genetic studies of complex traits.
• High-dimensional genomic data presents challenges for disease risk prediction, especially with limited sample sizes.
• Existing Polygenic Risk Score (PRS) methods have limitations in scalability or coefficient shrinkage.

Purpose of the Study:

• To introduce SSLPRS, a novel PRS method utilizing the Spike-and-Slab LASSO (SSL) prior.
• To develop a scalable inference algorithm for PRS using Genome-Wide Association Studies (GWAS) summary statistics.
• To evaluate the performance of SSLPRS compared to existing methods.

Main Methods:

• Developed a coordinate-ascent inference algorithm for SSLPRS operating on GWAS summary statistics.
• Utilized the Spike-and-Slab LASSO (SSL) prior for PRS inference.
• Validated the method through simulations and analysis of UK Biobank quantitative phenotypes.

Main Results:

• SSLPRS demonstrates competitive prediction accuracy and superior variable selection performance, particularly in sparse genetic architectures.
• Achieved over 50% improvement in positive predictive value in simulations.
• Selected variants in real phenotype analyses are enriched for significant genomic annotations and show improved replication rates.

Conclusions:

• SSLPRS provides a robust and scalable approach for disease risk prediction from genotype data.
• The method bridges theoretical frameworks of sparse Bayesian priors and penalized regression.
• SSLPRS offers enhanced variable selection and prediction accuracy, advancing polygenic risk score methodologies.