GEiPRS: a fast and powerful machine learning method for polygenic risk score prediction by leveraging...

Area of Science:

• Genetics
• Bioinformatics
• Machine Learning

Background:

• Penalized regression is standard for variant selection and polygenic risk score (PRS) analysis in genome-wide association studies (GWAS).
• Existing PRS methods often fail to incorporate genotype-environment interaction (GEI) and struggle with high-dimensional GWAS data.

Purpose of the Study:

• To propose a novel machine learning-based PRS method, GEiPRS, that simultaneously models genotype (G) and GEI effects.
• To efficiently handle high-dimensional GWAS data for variant selection, PRS construction, and prediction.

Main Methods:

• Developed GEiPRS, a machine learning approach incorporating G and GEI.
• Introduced Group ITerative LAsso with Batch Screening (GITLABS) for efficient variant selection and PRS construction.
• GITLABS employs strong rule screening, GL/SGL model fitting, and safe rule validation for computational efficiency.

Main Results:

• GEiPRS demonstrated superior performance over existing PRS methods in simulations.
• Outperformed in GEI-PRS association P-values, prediction accuracy, and subgroup risk stratification.
• Showcased enhanced computational efficiency for high-dimensional data.

Conclusions:

• GEiPRS effectively models GEI and handles high-dimensional GWAS data.
• The GITLABS algorithm provides an efficient solution for variant selection and PRS construction.
• GEiPRS shows promise for improved genetic risk prediction and subgroup analysis in large-scale datasets like the UK Biobank.