Deep Learning for Inferring Distribution of Time to the Last Common Ancestor from a Diploid Genome

Genomic data is a rich source of information about population history. In particular, for actively recombining species the time to the last common ancestor (LCA) between two chromosomes might be different in different chromosome loci. Estimating local LCA time is important for many problems: it can be used to infer genes under selection, or to infer effective population size changes. The current state-of-the art method PSMC to infer local LCA time and effective population size is based on a Hidden Markov Model. In this work we propose a new deep learning framework for local LCA time inference at the full genome scale. We demonstrate that our method is accurate in both local LCA time and, as a consequence, at the LCA time distribution which in turn translates into effective population size trajectory. In future our approach can be generalised for complex population scenarios.

Automated summary

This work proposes a new deep learning framework for inferring local last common ancestor (LCA) time at the full genome scale. The accuracy of the method in both local LCA time and LCA time distribution is demonstrated, which translates into effective population size trajectory.