| In this dissertation, composed of four independent chapters, I present high performance methods and computer software for molecular phylogenetic inference. Chapter 1 describes the Crux software package, which I developed to support the methods research in the subsequent chapters. Chapter 2 describes the relaxed neighbor joining algorithm, which is a far faster variant of the venerable neighbor joining algorithm. Chapter 3 applies a combination of mixture models and polytomous tree support, including several novel enhancements to existing algorithms, to call into question the validity of analyses that apply a single model of evolution to multi-gene concatenated data sets. Chapter 4 synthesizes the work of others to determine that approximated model probabilities can have a significant negative impact on decision-theoretic model selection. |
