A stochastic Dollo model for lateral transfer

Thesis

Lateral transfer, a process whereby species exchange evolutionary traits through non-ancestral relationships, is a frequent source of model misspecification in phylogenetic inference. Lateral transfer obscures the phylogenetic signal in the data — the signal of the taxa ancestry — as the histories of affected traits are mosaics of the species phylogeny and may conflict with the underlying phylogeny. We control for the effect of lateral transfer in a Stochastic Dollo model and a Bayesian setting. We infer rooted phylogenetic trees. Our likelihood is highly intractable as its parameters are given by the solution of a sequence of systems of differential equations which represent the expected evolution of traits along a tree and grow exponentially in dimension with the number of taxa under consideration. We construct an accurate parameter approximation framework, and from this we derive an efficient exact-approximate inference scheme. We illustrate our method on data sets of lexical traits in Eastern Polynesian and Indo-European languages and obtain improved fits over the corresponding model without lateral transfer.

Authors: Kelly, L

• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford