Negative selection on complex traits limits phenotype prediction accuracy between populations

Summary Phenotype prediction is a key goal for medical genetics. Unfortunately, most genome-wide association studies are done in European populations, which reduces the accuracy of predictions via polygenic scores non-European populations. Here, we use population genetic models to show that human demographic history and negative selection on complex traits can result population-specific architectures. For where alleles with largest effect trait under strongest selection, approximately half heritability be accounted by variants Europe absent from Africa, leading poor performance phenotype across these Further, such model, individuals tails risk distribution may not identified generated another population. We empirically test building model stratify between European-specific shared applied it 37 diseases UK Biobank. Across phenotypes, ?30% comes variants. conclude need include more diverse populations enable utility all