490,319 whole genomes improve rare-variant imputation

Researchers built a genotype-imputation reference panel from 490,319 UK Biobank whole genomes and released EagleImp plus its cloud implementation, EagleImp-RAP. Imputation uses a reference population to infer variants not directly measured in a study sample, so panel scale and diversity affect disease-association discovery. Compared with TOPMed r3, which contains 133,597 people, the new panel lowered phasing switch error in every global superpopulation except admixed American populations and in 25 of 26 1000 Genomes subpopulations. For variants down to 0.01% minor-allele frequency, mean absolute error improved in four of five superpopulations and 12 subpopulations. Reimputing European COVID-19 genome-wide association datasets added significant loci and sharpened fine-mapping. UK Biobank still does not represent the world, performance was not best for every American population, and privacy, ancestry bias and replication in outside cohorts remain constraints.