Improved polygenic prediction by Bayesian multiple regression on summary statistics.

Area of Science:

• Genomics
• Statistical Genetics
• Precision Medicine

Background:

• Predicting an individual's phenotype from DNA is a key goal in genomics.
• Existing methods for genomic prediction often require individual-level data or are computationally intensive.

Purpose of the Study:

• To develop and validate SBayesR, a novel Bayesian regression model utilizing genome-wide association studies (GWAS) summary statistics.
• To improve the accuracy and computational efficiency of genomic prediction compared to existing methods.

Main Methods:

• Extended the individual-level Bayesian multiple regression model (BayesR) to leverage GWAS summary statistics, creating SBayesR.
• Validated SBayesR using simulations, cross-validation on 12 real traits from UK Biobank data (1.1 million variants, 350,000 individuals).
• Compared SBayesR performance against LDpred and clumping/p-value thresholding methods using large GWAS meta-analysis summary statistics for height and BMI.

Main Results:

• SBayesR demonstrated improved prediction accuracy compared to state-of-the-art summary statistics methods.
• Achieved significant improvements in prediction R-squared: 5.2% over LDpred and 26.5% over clumping/p-value thresholding for height and BMI.
• SBayesR required a fraction of the computational resources compared to other methods.

Conclusions:

• SBayesR offers a computationally efficient and accurate method for genomic prediction using GWAS summary statistics.
• This advancement has significant implications for precision medicine and large-scale genetic studies.
• The model effectively utilizes publicly available GWAS data for improved phenotype prediction.