84
practice to repeat the analysis in PhyML with randomized starting trees; if the same tree topology is recovered each time, it is
more likely that the globally optimal tree has been recovered.
11. Long branches on a phylogenetic tree arise from the inclusion
of highly divergent sequences and cause a systematic error
known as long branch attraction. If possible, long branches
should be broken up by including additional sequences that
have higher identity to the divergent sequences; otherwise,
the divergent sequences may need to be removed.
12. Phylogenetic trees inferred using maximum-likelihood can be
validated in several ways. First, the topology of the tree should
be robust to variations in the evolutionary models used in the
calculation; the phylogenetic analysis should be repeated using
alternative substitution matrices that scored highly in
Subheading 3.3, and the resulting tree topologies should be
very similar to the maximum-likelihood tree. Second, statistical support for individual branches in the tree should be
assessed using the bootstrap method, which involves repeating
the phylogenetic analysis many times (usually at least 100)
using multiple sequence alignments generated by random
resampling of the complete alignment, to determine whether
the same tree topology can be reproduced even when some of
the data are removed or duplicated. The bootstrap value of a
branch is the proportion of the bootstrap trees that contain
the same bifurcation of sequences as denoted by that branch.
Branches with bootstrap values <70% should be viewed with
suspicion. Finally, if the tree describes the evolution of orthologous sequences, it should be consistent with established
species- based trees (although this criterion is less applicable to
SBPs due to the prevalence of horizontal gene transfer).
13. In our experience, the file formatting requirements of PAML
are the most common cause of any errors encountered while
running the program. The alignment must be in PHYLIP
interleaved format. This format must be specified by adding
the letter “I” (for “interleaved”) to the end of the first line
(after the number of sequences and number of columns).
There must also be two spaces between each sequence name
and the corresponding sequence. The tree must be in Newick
format. PAML can only parse trees obtained from PhyML if
they do not contain branch support values. Trees with branch
support values can be reformatted using the program Retree
in the PHYLIP package; open the tree in Retree and write the
(unrooted) tree to a new file without further modification.
14. Ideally, multiple ancestral nodes between the reference SBP
and the last common ancestor of the sequences of interest
would be selected and characterized experimentally.
Ben E. Clifton et al.
practice to repeat the analysis in PhyML with randomized starting trees; if the same tree topology is recovered each time, it is
more likely that the globally optimal tree has been recovered.
11. Long branches on a phylogenetic tree arise from the inclusion
of highly divergent sequences and cause a systematic error
known as long branch attraction. If possible, long branches
should be broken up by including additional sequences that
have higher identity to the divergent sequences; otherwise,
the divergent sequences may need to be removed.
12. Phylogenetic trees inferred using maximum-likelihood can be
validated in several ways. First, the topology of the tree should
be robust to variations in the evolutionary models used in the
calculation; the phylogenetic analysis should be repeated using
alternative substitution matrices that scored highly in
Subheading 3.3, and the resulting tree topologies should be
very similar to the maximum-likelihood tree. Second, statistical support for individual branches in the tree should be
assessed using the bootstrap method, which involves repeating
the phylogenetic analysis many times (usually at least 100)
using multiple sequence alignments generated by random
resampling of the complete alignment, to determine whether
the same tree topology can be reproduced even when some of
the data are removed or duplicated. The bootstrap value of a
branch is the proportion of the bootstrap trees that contain
the same bifurcation of sequences as denoted by that branch.
Branches with bootstrap values <70% should be viewed with
suspicion. Finally, if the tree describes the evolution of orthologous sequences, it should be consistent with established
species- based trees (although this criterion is less applicable to
SBPs due to the prevalence of horizontal gene transfer).
13. In our experience, the file formatting requirements of PAML
are the most common cause of any errors encountered while
running the program. The alignment must be in PHYLIP
interleaved format. This format must be specified by adding
the letter “I” (for “interleaved”) to the end of the first line
(after the number of sequences and number of columns).
There must also be two spaces between each sequence name
and the corresponding sequence. The tree must be in Newick
format. PAML can only parse trees obtained from PhyML if
they do not contain branch support values. Trees with branch
support values can be reformatted using the program Retree
in the PHYLIP package; open the tree in Retree and write the
(unrooted) tree to a new file without further modification.
14. Ideally, multiple ancestral nodes between the reference SBP
and the last common ancestor of the sequences of interest
would be selected and characterized experimentally.
Ben E. Clifton et al.