| Mark Holder email at ku.edu with the username: mtholder Department of Ecology and Evolution 6031 Haworth University of Kansas 1200 Sunnyside Ave Lawrence KS 66045 |
My research is focused on phylogenetic inference. I am interested in both the statistical basis of tree estimation and the algorithms used to explore phylogenies.
I joined the Department of Ecology and Evolution at the University of Kansas as an assistant professor in August, 2007.
My CV in pdf is available here, and bibliographic data for my publications is contained in this BibTeX file.
I post my calendar here.
Information about a postdoctoral position available in my lab can be found here.
My current work is a mixture of developing methods of tree estimation and implementing them in software.
In Bayesian phylogenetics, we learn about the evolutionary tree for a group of taxa by accounting for both our prior beliefs and information from the data (via the likelihood under a specific model). For virtually all data sets, we must approximate the solution using a computational tool called Markov chain Monte Carlo (MCMC). Adopting a Bayesian perspective is attractive for several reasons:
Paul Lewis, Kent Holsinger, and I have done research on using prior probabilities that allow for polytomies in phylogenies. This can reduce the incidence of incorrectly inferred relationships that receive strong support in terms of posterior probability. I am still investigating the subject of robust prior distributions for trees. The prior probabilities on parameters of the model of character evolution can also affect the posterior probability of clades. Derrick Zwickl and I collaborated on an exploration of when vague priors on the GTR model can lead to poor performance.
The mechanics of the MCMC algorithm itself should not affect the final conclusion of a Bayesian analysis, but these details can have profound effects on the efficiency of the procedure. I am interested in developing new proposals (also referred to as "moves") for phylogenetic MCMC.
References:
The answers to many evolutionary and ecological questions depend on an underlying phylogeny, leading to an explosion in the use of comparative methods and character mapping. Recently I have developed an algorithm to speed up the process of finding the joint maximum likelihood estimate of the ancestral state reconstruction of a character on a phylogeny. This algorithm has been implemented in LASRDisc by Vanessa Jackson.
I am also very interested in developing models to incorporate phylogeny into specific problems in biology. While I was working with Paul Lewis at the University of Connecticut, we worked with several ecologists and statisticians to use phylogenetic information to improve the explanation of the distributions of Protea species. The phylogeny is not a major determinant of the distribution of these species, but a failure to consider the phylogeny at all will produce artifacts that affect the ecological model being used.
References:
I am a member of the CIPRES project which is geared toward the development of informatics tools for inference evolutionary trees for thousands of taxa. Traditional approaches to estimating phylogenies perform very poorly on trees of this size. The initial release of the software will implement recursive, iterative, Divide-and-Conquer-Method-3 (rec-I-DCM3), one of the divide-and-conquer approaches developed by Tandy Warnow and collaborators. We are working to improve these approaches in terms of both speed and accuracy.
My dissertation work used simulations of sequence data under a complex model of evolution to compare the performance of distance, parsimony, and maximum likelihood strategies for inferring trees. The goal of the work was to examine performance when several important assumptions of the analyses are violated simultaneously. I am expanding these performance tests to include tests of Bayesian approaches. These results will be of particular interest because Bayesian methods (or at least the posterior probabilities that summarize the strength of support in Bayesian analyses) may be more sensitive to model violation than using maximum likelihood with non-parametric bootstrapping.
I am writing a manuscript on the statistical consistency conditions for Paul Lewis's Mk model that can be applied to morphological data.
Paul Lewis, David Swofford, and I are preparing a manuscript synthesizing work on the statistical basis of parsimony.
Before coming to KU, I am worked with David Swofford at Florida State University in a postdoctoral position funded by the CIPRES project. Before that I was a postdoc with Paul Lewis at the University of Connecticut and a graduate student of David Hillis at the University of Texas.