Comparative evaluation of maximum parsimony and Bayesian phylogenetic reconstruction using empirical morphological data. (18th July 2018)

• Record Type: Journal Article
• Authors: Schrago, Carlos G.; Aguiar, Barbara O.; Mello, Beatriz
• Is Part Of: Journal of evolutionary biology. Volume 31:Number 10(2018)
• Page Start: 1477
• Page End: 1484
• Publication Date: 2018-07-18
• Subjects: evolutionary model -- Mk model -- MorphoBank -- morphology evolution -- simulation; Evolution (Biology) -- Periodicals; Biology -- Periodicals; 576.8
• DOI: 10.1111/jeb.13344
• Languages: English
• ISSNs: 1010-061X

Abstract

Abstract: The use of discrete morphological data in Bayesian phylogenetics has increased significantly over the last years with the proposal of total evidence analysis and the treatment of fossils as terminal taxa in Bayesian molecular dating. Both approaches rely on the assumption that probabilistic Markov models reasonably accommodate all the complexity of morphological evolution of discrete traits. The performance of such morphological models used in Bayesian phylogenetics has been thoroughly investigated, but conclusions so far were based mostly on simulated data. In this study, we have surveyed MorphoBank and obtained a large number of morphological matrices to evaluate Bayesian phylogenetic inference (BI) under Lewis' Mk model in comparison with the maximum parsimony (MP) algorithm. We found that trees estimated by both methods frequently differed and that BI generated a larger amount of polytomic tree topologies. The number of trees contained in the 95% Bayesian credibility interval was significantly greater than the number of equally parsimonious trees. We also investigated which factors mostly influenced the topological difference between maximum parsimony and Bayesian tree topologies and found that the number of terminals in morphological matrices was the variable with the highest association with the topological distance between trees inferred by BI and MP. Surprisingly, we show that differences between both approaches were not influenced by increasing sample size. Our results, which were based on a large set of empirical matrices, corroborate recent findings that BI is less precise than MP.