Approaches to Ancestral Pangenomes and Their Phylogenetic Reconstruction

Abstract

We investigate the problem of reconstructing ancestral pangenomes from present- day genomic data by modelling structural variation and evolutionary turnover. Our first chapter models ancestral and descendant pangenomes as sets of gene adjacencies, using phylogenetic validation - based on Dollo's law - to filter out recent innovations unlikely to have existed in any ancestor. This approach enables a meaningful reconstruction of ancestral gene order via iterative steinerization, even without optimization. In a second chapter, we complement this framework with a probabilistic tree model in which discrete objects (e.g., genes or features) are transmitted down a hierarchical phylogeny and replaced, respecting Dollo, with a certain probability. We derive theoretical expectations for the retention and overlap of ancestral objects across nodes and assess the accuracy of steinerization-based reconstruction in simulated datasets. Our simulations demonstrate that while theoretical predictions align with observed retention under low replacement rates, random divergence among novel objects introduces noise in deeper or faster-evolving trees. Together, these studies provide some promising approaches to understanding the limits and potential of ancestral reconstruction in a pangenomic landscape.