Transferability of polygenic risk scores depending on demography and dominance coefficients
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
The genetic liability to a complex phenotype is calculated as the sum of genotypes, weighted by effect size estimates derived from summary statistics of genome-wide association study (GWAS) data. Due to different allele frequencies (AF) and linkage disequilibrium (LD) patterns across populations, polygenic risk scores (PRS) that were developed on one population drop drastically in predictive performance when transferred to another. One of the major factors contributing to AF and LD heterogeneity is genetic drift, which acts strongly during population bottlenecks and is influenced by the dominance of certain alleles. In particular, since the causal variants on empirical data are typically not known, the presence of population specific LD-patterns will strongly affect transferability of PRS models. In this work, we therefore conducted demographic simulations to investigate the influence of the dominance coefficient on the transferability of PRS among European, African and Asian populations. By modifying the length and size of the bottleneck leading to the split of Eurasian and African populations, we gain a deeper understanding of the underlying dynamics. Finally, we illustrate that PRS models that are adapted to the underlying dominance coefficient can substantially increase their prediction performance in out-of-target populations.
Polygenic risk scores (PRS) are increasingly used in clinical care for the management of many complex disorders such as breast cancer or cardiovascular diseases. Since heritability should be independent of ancestry so should be the predictability of the models. This is, however, currently not the case and the missing transferability of PRS is favoring individuals from European descent, who represent the largest population to train PRS. In this work we study on simulated populations what degree of transferability is theoretically achievable under different demographic models and dominance coefficients of the pathogenic variants. The results of our work are twofold: the effect of genetic drift and selection on the transferability can be quantified in simulations and recessive traits are more conserved.
