An estimator of first coalescent time reveals selection on young variants and large heterogeneity in rare allele ages among human populations

Allele age has long been a focus of population genetic research, primarily because it can be an important clue to the fitness effects of an allele. By virtue of their effects on fitness, alleles under directional selection are expected to be younger than neutral alleles of the same frequency. We developed a new coalescent-based estimator of a close proxy for allele age, the time when a copy of an allele first shares common ancestry with other chromosomes in a sample not carrying that allele. The estimator performs well, including for the very rarest of alleles that occur just once in a sample, with a bias that is typically negative. The estimator is mostly insensitive to population demography and to factors that can arise in population genomic pipelines, including the statistical phasing of chromosomes. Applications to 1000 Genomes Data and UK10K genome data confirm predictions that singleton alleles that alter proteins are significantly younger than those that do not, with a greater difference in the larger UK10K dataset, as expected. The 1000 Genomes populations varied markedly in their distributions for singleton allele ages, suggesting that these distributions can be used to inform models of demographic history, including recent events that are only revealed by their impacts on the ages of very rare alleles.Author summary: We developed a way to estimate the time when a copy of a gene most recently shared ancestry with other copies of that gene. This is also an estimate of the upper bound of when a mutation has arisen, and it can be used to study the ages of alleles that are found in a population. The method can be applied to the very rarest alleles found only once in a sample, even in studies of many thousands of genomes. We tested the method extensively, found it performs well, and can be used under a wide variety of conditions. We applied it to 1000 Genomes project data (26 populations) and the UK10K data (over 7000 genomes) and found clear evidence that alleles that change proteins are younger than alleles that do not, as expected. We also observed wide variation in the ages of alleles at low frequency among the 1000 Genome project populations, indicating that our method could be used to study the demographic history of human populations. Going forward, the estimator should be useful for many kinds of questions in population genomics, particularly as sample sizes continue to grow.