Improving the computation efficiency of polygenic risk score modeling: faster in Julia

Abstract

To enable large-scale application of polygenic risk scores (PRSs) in a computationally efficient manner, we translate a widely used PRS construction method, PRS-continuous shrinkage, to the Julia programming language, PRS.jl. On nine different traits with varying genetic architectures, we demonstrate that PRS.jl maintains accuracy of prediction while decreasing the average runtime by 5.5×. Additional programmatic modifications improve usability and robustness. This freely available software substantially improves work flow and democratizes usage of PRSs by lowering the computational burden of the PRS-continuous shrinkage method.