Multiple sources of uncertainty confound inference of historical human generation times

Abstract

WO_SCPLOWANGC_SCPLOW et al. (2023) recently proposed an approach to infer the history of human generation intervals from changes in mutation profiles over time. As the relative proportions of different mutation types depend on the ages of parents, binning variants by the time they arose allows for the inference of average paternal and maternal generation intervals over times. Applying this approach to published allele age estimates, WO_SCPLOWANGC_SCPLOW et al. (2023) inferred long-lasting sex differences in average generation times and surprisingly found that ancestral generation times of West African populations remained substantially higher than those of Eurasian populations extending tens of thousands of generations into the past. Here we argue that the results and interpretations in WO_SCPLOWANGC_SCPLOW et al. (2023) are primarily driven by noise and biases in input data and a lack of validation using independent approaches for estimating allele ages. With the recent development of methods to reconstruct genome-wide gene genealogies, coalescence times, and allele ages, we caution that downstream analyses may be strongly influenced by uncharacterized biases in their output.