Polygenic scores via penalized regression on summary statistics.

Area of Science:

• Genetics
• Biostatistics
• Epidemiology

Background:

• Polygenic scores (PGS) aggregate genetic variants to predict disease risk or phenotypes.
• Current methods for calculating PGS often struggle to incorporate linkage disequilibrium (LD) information from external sources, especially when using summary statistics.

Purpose of the Study:

• To propose a novel method, lassosum, for constructing PGS that effectively utilizes summary statistics and external LD information.
• To introduce a general method for selecting the tuning parameter for lassosum in the absence of validation data.

Main Methods:

• Developed lassosum, a penalized regression framework for PGS construction using summary statistics and a reference panel.
• Employed pseudovalidation for tuning parameter selection, comparing its effectiveness against using validation data and a conservative default setting.

Main Results:

• Simulations demonstrated that pseudovalidation yields prediction accuracy comparable to using validation data and superior to a conservative tuning parameter choice.
• Lassosum significantly outperformed simple clumping and P-value thresholding methods in prediction accuracy across various scenarios.
• Lassosum showed substantial improvements in speed and accuracy compared to the LDpred method.

Conclusions:

• Lassosum provides a robust and efficient method for calculating polygenic scores by integrating summary statistics with linkage disequilibrium information.
• The proposed pseudovalidation technique offers a reliable approach for tuning parameter selection, enhancing the practical utility of lassosum.
• Lassosum represents a significant advancement in genetic risk prediction, offering improved accuracy and computational efficiency over existing methodologies.