Novel Phylogenetic Methods: Supertrees and Phylogenetic Diversity

This thesis considers novel methods in two aspects of phylogenetics: supertrees and phylogenetic diversity. In each area, I examine a recently described novel method, explore it’s properties, expand it, compare it’s performance to a popular established method using simulated data and develop new tests. Supertree methods construct a tree from a set of input trees with overlapping taxa. This thesis focuses on the maximum likelihood supertree method. Open challenges with the method are addressed, including the implications of the normalising constant on the choice of distance metric and the inclusion of non-binary input trees. An alternative likelihood model, a new method - the maximum likelihood consensus method - and a diagnostic test for evaluating the fit of the likelihood model and the performance of the method are proposed. Sets of input trees are generated under different conditions, including incomplete lineage sorting, and model misspecification. The performance of the methods differs between the simulations, but is found to be dependent upon the level of variation between the input trees. The maximum likelihood supertree method is as effective as a popular method, matrix representation with parsimony (MRP). The maximum likelihood consensus method is as effective as MRP and majority-rule consensus. Phylogenetic trees can be used in conservation to assess the diversity of a set of species and select the set that maximises diversity. This thesis focuses on the maximum minimum distance (MMD) method. Using feature alignments generated by a birth-death Dollo parsimony model, MMD is compared to phylogenetic diversity (PD). As the variation of the rate of evolution along the tree increases and/or the feature loss rate increases, MMD is generally increasingly more successful than PD. However, predicting whether MMD or PD will be more successful for a given tree is a difficult task. MMD also appears to be more robust to errors in the tree or distance matrix estimation than PD. A new method, Maximising Diversity with Feature Loss (MDFL), is proposed, which is as at least as successful as both PD and MMD in these simulations. The properties of MMD and MDFL are explored and some extensions proposed.