Penalized regression and model selection methods for polygenic scores on summary statistics.

Area of Science:

• Genetics
• Biostatistics
• Computational Biology

Background:

• Polygenic scores are crucial for predicting complex disease risk.
• Developing polygenic scores from summary statistics is an active research area.

Purpose of the Study:

• To propose a novel penalized regression method for constructing polygenic risk scores using summary statistic data.
• To extend existing methods by incorporating the Truncated Lasso Penalty (TLP) and elastic net.

Main Methods:

• Utilized penalized regression with summary statistic and reference data.
• Extended the LassoSum framework to include Truncated Lasso Penalty (TLP) and elastic net.
• Developed methods for approximating AIC and BIC for model selection without validation data.
• Proposed a quasi-correlation metric for evaluating out-of-sample predictive accuracy.

Main Results:

• The TLP method demonstrated improved predictive accuracy and sparsity compared to LASSO in simulations and real data.
• Proposed methods facilitate model selection and evaluation of polygenic risk scores using only summary statistics.
• The approach was successfully applied to Genome-Wide Association Studies (GWAS) for lipids, height, and lung cancer.

Conclusions:

• The proposed penalized regression methods enhance the construction and application of polygenic risk scores from summary statistics.
• These methods improve predictive accuracy and enable robust model selection and evaluation in the absence of validation data.
• The TLP approach offers a valuable tool for genetic risk prediction in complex diseases.