Create your own conference schedule! Click here for full instructions

Abstract Detail

Genomics / Proteomics

Clements, Mark [1], Moore, Michael [2], Mathews, Sarah [1].

Mixture models versus a priori partitioning schemes in analyses of plastome data sets: alarge angiosperm matrix revisited.

Our understanding of relationships among major clades of angiosperms have been substantially improved by analyses of plastome data sets that include the majority of protein-coding genes found in the plastid genome. However, uncertainty remains around some nodes, despite the large number of genes (up to 83) and characters (up to ~66,700) in these matrices. We explored a previously published 83-gene 94-taxon angiosperm plastome data set, using mixture model approaches available in BayesPhylogenies and PhyloBayes to analyze both nucleotide and amino acid matrices under several different assumptions about models of evolution. We found that 7 of the 13 nodes that were poorly supported in the published analyses were better supported in the mixture model analyses, including the node that attaches Ceratophyllum to the eudicots. In several cases, support was substantially increased. Conversely, we found several nodes where the data appear inadequate to yield robust resolution. For example our results indicate that there is more uncertainty around some nodes at the base of the eudicots and within asterids and rosids than is indicated by the published support values. In most cases, results from nucleotide and amino acid searches were similar if not identical. A notable exception was the attachment of Chloranthaceae to the rest of the angiosperms just above the ANITA grade in the amino acid trees (posterior probability of 0.99). We found no differences between results from searches that did or did not account for heterotachy. Together the results suggest that improved support resulted largely from the use of mixture models, possibly because the codon and gene partitioning schemes used in the published analyses actually are a poor fit to the data. Our mixture model searches found the optimal number of partitions, based on pattern heterogeneity, to be between 13 and 19, many fewer than the number of genes but many more than the number of codon positions.

Broader Impacts:

Log in to add this item to your schedule

1 - Harvard University, Arnold Arboretum, 1300 Centre Street, Boston, MA, 02131, USA
2 - Oberlin College, Department of Biology, 119 Woodland St., Oberlin, OH, 44074, USA

data partitioning
mixture models
Plastid genome.

Presentation Type: Oral Paper:Papers for Topics
Session: 44
Location: Union B/Hyatt
Date: Wednesday, July 11th, 2012
Time: 8:45 AM
Number: 44004
Abstract ID:893

Copyright 2000-2012, Botanical Society of America. All rights reserved