COMPARING AND CONTRASTING PARSIMONY AND MAXIMUM LIKELIHOOD APPROACHES TO PALEONTOLOGICAL PHYLOGENETICS USING TRILOBITES AS A MODEL SYSTEM

Traditionally, paleontologists have used parsimony analysis to estimate tree topology in their analyses of fossil taxa. However, complications do arise when applying parsimony due to factors such as missing data, long-branch attraction, and large clades whose taxonomic diversity exceeds the number of characters available for phylogenetic analysis; these can affect results and may limit resolution, suggesting that expanding the tool kit available to phylogenetically-minded paleontologists beyond the simple application of parsimony may lead to advances in phylogeny reconstruction. The use of maximum likelihood (ML) may help obviate some of these complications, but naïve application of the same ML methods used for molecules to morphology is problematic because of several differences between morphological and molecular data. In particular, modifications to likelihood models have been developed by Paul Lewis and others that accommodate some of the ascertainment biases implicit in morphological phylogenetic approaches that make these types of data so different from molecular data. These ascertainment biases include only sampling variable data or excluding autapomorphies. We have extended the GARLI software by implementing models that included extensions of Lewis' approach for correcting for ascertainment biases. These extension enable ML tree searching for fossil data sets. These modifications make the application of ML to phylogeny estimation with morphological data more tractable. Here we apply ML methods to phylogenetic analysis of an enigmatic collection of Ordovician cheirurid trilobites previously referred to the Eccoptochilinae. We especially focus on the differences between results from parsimony and ML, although each supports a paraphyletic “Eccoptochilinae” and suggests the group is nested within another subfamily, the Sphaerexochinae. We also consider differences between ML analyses of groups that differ substantially in size, using a previous analysis of Early Cambrian olenelloid trilobites as a comparison. ML phylogenetic methods applied to fossil taxa appear to provide great promise, but at this time analyses require substantial computational time. Further, thus far with large clades the degree of resolution is less than that attained by parsimony-based supertree approaches.