Deep estimation of the intensity and timing of selection from ancient genomes

Leveraging past allele frequencies has proven to be key to identify the impact of natural selection across time. However, this approach often suffers from imprecise estimations of the intensity (s) and timing (T) of selection particularly when ancient samples are scarce in specific epochs. Here, we aimed at bypassing the computation of past allele frequencies by implementing new convolutional neural networks (CNNs) algorithms that directly use ancient genotypes sampled across time to refine the estimations of selection parameters. Using computer simulations, we first show that genotype-based CNNs consistently outperform an approximate Bayesian computation (ABC) approach based on past allele frequency trajectories, regardless of the selection model assumed and of the amount of ancient genotypes available. When applying this method to empirical data from modern and ancient Europeans, we confirmed the reported excess of selection events in post-Neolithic Europe, independently of the continental subregion studied. Furthermore, we substantially refined the ABC-based estimations of s and T for a set of positively-and negatively-selected variants recently identified, including iconic cases of positive selection and experimentally validated disease-risk variants. Thanks to our CNN predictions we provide support to the history of recent and strong selection in northern Europe associated to the Black Death pandemic and confirm the heavy burden recently imposed by tuberculosis in Europe. These findings collectively support that detecting the imprints of natural selection on ancient genomes are crucial for unraveling the past history of severe human diseases.