|
|
|
| Abstract Title:
|
| To combine or not to combine: An experimental study comparing the accuracy of supertree and combined analysis methods in the reconstruction of large phylogenies.
|
| Graduate Student Presenter:
|
M. Shel Swenson
|
| Name of the Author(s) and Affiliation(s):
|
M. Shel Swenson, Deptartment of Mathematics, The University of Texas at Austin; Francois Barbancon, Microsoft, Redmond, WA; C. Randal Linder, Section of Integrative Biology, UT-Austin; Tandy Warnow, Deptartment of Computer Science, UT-Austin
|
|
Supertree methods comprise one approach to reconstructing large phylogenies, whereby trees are estimated for subsets of an entire set of taxa, and then combined into a single tree on that full set of taxa using some algorithmic techniques (such as matrix representation with parsimony, or MRP). The competing approach is to perform a combined analysis (also known as a "super-matrix" or "total evidence" approach), whereby the different data matrices for each of the different subsets of taxa are put into a single matrix, and a tree is estimated on that "super-matrix". A previous simulation study by Bininda-Emonds and Sanderson suggested that supertree methods would perform better than combined analyses with respect to the topological accuracy of the resultant phylogenetic tree. The experimental design of that study, however, did not take into consideration several biological and systematic factors, leaving open the question of how well these different approaches would actually perform. In our study, we designed mathematical models of gene birth and death and of data sampling practice in systematics. Using these models, we performed an extensive simulation study and have found that, in contrast to results of the Bininda-Emonds and Sanderson 2001 simulation study, that the super-matrix approach definitively outperforms supertree methods, with the supertree methods approaching the accuracy of super-matrix methods only under unusual and highly restrictive conditions.
|
|
|
