Link to bioRxiv paper:

http://biorxiv.org/cgi/content/short/2020.07.24.220392v1?rss=1

Authors: Reales, G., Kelemen, M., Wallace, C.

Abstract:

Polygenic scores (PGS) aim to predict complex traits at an individual level based on genetic data. Computation of PGS is based on simply ascertained genome-wide association summary statistics but typically requires an independent test dataset to tune PGS parameters, and for the more sophisticated methods, the computation of LD matrices. Internally tuned methods have recently been proposed that obviate the need for a test dataset, but they remain computationally intensive to run. Here we present RapidoPGS, a flexible and fast method to compute PGS without the need to compute LD-matrices, requiring only summary-level GWAS datasets. Based on fine-mapping principles, RapidoPGS computes the posterior probability that each variant is causal, which in turn is used to shrink effect sizes adaptively as a function of LD and strength of association. We show by summary and individual-level validation that RapidoPGS performs well in comparison with another well-established internally-trained method (median AUC difference [RapidoPGS - LDpred2-auto] = -0.02205, range = [-0.1184, 0.0217]), with at least 50-fold improved speed. RapidoPGS is implemented in R, and can work with user-supplied summary statistics or download them directly from the GWAS catalog. We propose that RapidoPGS can be used to rapidly screen a set of candidate traits for utility, before more computationally intensive methods are applied to selected traits.

Copy rights belong to original authors. Visit the link for more info