6. Phylogenetic Analyses of Large Data Sets
99
ments in the analysis of large data sets is a new computer program that greatly
enhances the ability to find shorter trees using parsimony. K. Nixon (unpubI.) has
developed a new method called the RATCHET, which is applied using his program
DADA along with P. Goloboff's program NONA.
When applied to the 567-taxon matrix of three genes for angiosperms (analyses
conducted by K. Nixon), the RATCHET quickly found trees shorter than those
recovered by P AUP* 4.0 (Swofford 1998). After several weeks, searches with P AUP*
4.0 found trees of 45,107 steps whereas the RATCHET found trees of length 45,101
steps (that is, six steps shorter than the trees found using P AUP*) in only a few
hours, and subsequent analyses found trees of 45,100 steps. In addition to being
fast, the RATCHET method makes it possible to consider very large numbers of
trees from multiple islands, decreasing one's chances of being marooned on a single
island of trees and improving the reliability of the consensus of the shortest trees.
For example, with the 567-taxon, three-gene data set, PAUP* found approximately
1600 trees of length 45,107, all on one island over the course of several weeks,
whereas the RATCHET recovered nearly 5000 trees of length 45,100 in approximately 24 hours.
The phylogenetic analysis of large data sets via parsimony has benefited greatly
from recent developments in software; continued developments in this area are
anticipated, making the analysis of large data sets ever more straightforward.
6 Other Approaches: Super Trees
Another approach that permits the construction of trees for large numbers of taxa is
the "supertree" method (Sanderson et ai. 1998). Using this approach, existing topologies that overlap in at least several taxa are "grafted" together (Sanderson et ai.
1998). Although this approach may have some applications in the study of large
numbers of taxa, it cannot, we believe, substitute for thc actual analysis of the large
number of taxa needed to infer phylogeny accurately across large groups. The
supertree method has potential difficulties and shortcomings, some of which are
reviewed by Wilkinson and Thorley (1998). Perhaps the biggest problem with the
supertree approach is that it reconstructs a phylogeny across a diverse group, such
as the angiosperms, via a piecemeal approach where only one or a few taxa may be
shared among trees. To assess phylogeny across the angiosperms, a group for which
the spine of the tree has been uncertain, as have the major clades, only a global
approach would yield meaningful results.
There are, however, potential applications of a "tree grafting" approach when
used in combination with the methods of analysis outlined -in the sections above.
That is, once a large data set has been analyzed phylogenetically and an overall
topology obtained, it may be useful to graft existing clades that have been the subject of more thorough sampling onto the main tree. For example, using the threegene topology for angiosperms (Soltis et aI., submitted), it would seem appropriate
to graft well-sampled and studied clades onto the overall tree to replace some of the
99
ments in the analysis of large data sets is a new computer program that greatly
enhances the ability to find shorter trees using parsimony. K. Nixon (unpubI.) has
developed a new method called the RATCHET, which is applied using his program
DADA along with P. Goloboff's program NONA.
When applied to the 567-taxon matrix of three genes for angiosperms (analyses
conducted by K. Nixon), the RATCHET quickly found trees shorter than those
recovered by P AUP* 4.0 (Swofford 1998). After several weeks, searches with P AUP*
4.0 found trees of 45,107 steps whereas the RATCHET found trees of length 45,101
steps (that is, six steps shorter than the trees found using P AUP*) in only a few
hours, and subsequent analyses found trees of 45,100 steps. In addition to being
fast, the RATCHET method makes it possible to consider very large numbers of
trees from multiple islands, decreasing one's chances of being marooned on a single
island of trees and improving the reliability of the consensus of the shortest trees.
For example, with the 567-taxon, three-gene data set, PAUP* found approximately
1600 trees of length 45,107, all on one island over the course of several weeks,
whereas the RATCHET recovered nearly 5000 trees of length 45,100 in approximately 24 hours.
The phylogenetic analysis of large data sets via parsimony has benefited greatly
from recent developments in software; continued developments in this area are
anticipated, making the analysis of large data sets ever more straightforward.
6 Other Approaches: Super Trees
Another approach that permits the construction of trees for large numbers of taxa is
the "supertree" method (Sanderson et ai. 1998). Using this approach, existing topologies that overlap in at least several taxa are "grafted" together (Sanderson et ai.
1998). Although this approach may have some applications in the study of large
numbers of taxa, it cannot, we believe, substitute for thc actual analysis of the large
number of taxa needed to infer phylogeny accurately across large groups. The
supertree method has potential difficulties and shortcomings, some of which are
reviewed by Wilkinson and Thorley (1998). Perhaps the biggest problem with the
supertree approach is that it reconstructs a phylogeny across a diverse group, such
as the angiosperms, via a piecemeal approach where only one or a few taxa may be
shared among trees. To assess phylogeny across the angiosperms, a group for which
the spine of the tree has been uncertain, as have the major clades, only a global
approach would yield meaningful results.
There are, however, potential applications of a "tree grafting" approach when
used in combination with the methods of analysis outlined -in the sections above.
That is, once a large data set has been analyzed phylogenetically and an overall
topology obtained, it may be useful to graft existing clades that have been the subject of more thorough sampling onto the main tree. For example, using the threegene topology for angiosperms (Soltis et aI., submitted), it would seem appropriate
to graft well-sampled and studied clades onto the overall tree to replace some of the