Identifying a species tree subject to random lateral gene transfer

TL;DR

This paper analyzes the challenges of reconstructing species trees in the presence of lateral gene transfer (LGT), showing that triplet-based methods can be consistent under certain conditions and introducing a novel connection to random walks on cyclic graphs.

Contribution

It provides an exact analysis of species tree reconstruction under LGT, identifying conditions for consistency and proposing a new method with practical applications.

Findings

Triplet-based methods fail in some LGT scenarios.

Consistency is achievable if LGT transfer rate is low.

A novel link between LGT and random walks on cyclic graphs is established.

Abstract

A major problem for inferring species trees from gene trees is that evolutionary processes can sometimes favour gene tree topologies that conflict with an underlying species tree. In the case of incomplete lineage sorting, this phenomenon has recently been well-studied, and some elegant solutions for species tree reconstruction have been proposed. One particularly simple and statistically consistent estimator of the species tree under incomplete lineage sorting is to combine three-taxon analyses, which are phylogenetically robust to incomplete lineage sorting. In this paper, we consider whether such an approach will also work under lateral gene transfer (LGT). By providing an exact analysis of some cases of this model, we show that there is a zone of inconsistency for triplet-based species tree reconstruction under LGT. However, a triplet-based approach will consistently reconstruct a species tree under models of LGT, provided that the expected number of LGT transfers is not too high. Our analysis involves a novel connection between the LGT problem and random walks on cyclic graphs. We have implemented a procedure for reconstructing trees subject to LGT or lineage sorting in settings where taxon coverage may be patchy and illustrate its use on two sample data sets.