In this manuscript we answer questions about the predictive performance of polygenic scores. Based on ancestry, we find large statistically significant differences in performance between populations. A second major area of inquiry concerns differences in the distributions of polygenic scores for worldwide populations, as currently calculated30–39. (The term “worldwide populations” is used in this manuscript. It is important to note that there are many meanings and operationalizations of the term “population”, and that only some of the world’s populations are represented in genetic analyses). Multiple potential causes of observed distribution differences of polygenic scores have been reported, including drift32, selection33,36–39, artifactual differences due to uncorrected population stratification30,31, and different environmental effects40,41. We show that investigator-driven choices in the construction of polygenic scores also significantly impacts distributions of these polygenic scores for worldwide populations. Finally, in contrast to putative claims about the effect (or lack thereof) of polygenic score differences on phenotypic differences among worldwide populations, we show that current knowledge is insufficient to either confirm or refute such reports. These results calibrate researcher expectations about polygenic score performance in different populations and demonstrate that treatment of ancestry depends critically on how polygenic scores are constructed, for diverse non-European ancestry populations.
