Phylogeny-aware alignment with PRANK - Multiple Sequence Alignment Methods - page 160

with caution in analyses where the guide phylogeny is unknown prior
to alignment. It is possible that the alignment generated by PRANK is
influenced by the guide phylogeny and, in the case that the guide
phylogeny is seriously wrong, the subsequent phylogenetic analysis
based on the resulting alignment may be biased. On the other hand, if
the problem with the guide phylogeny can be sorted out, PRANK is
expected to provide superior alignments for evolutionary analyses,
often closely approximating alignments produced with computation-
ally much heavier statistical methods [ 14 , 15 ].
Co-estimation of alignment and phylogeny with an iterative
approach is a promising new idea [ 9 ] that can greatly reduce the
problems caused by the inter-dependency of the two inferences.
PRANK can merge two alignments using the very same phylogeny-
aware approach that is used for de novo alignment and the method
is in principle ready to be embedded in a similar framework,
splitting the alignment task to speed up the search. We are in the
process of studying the best practices to divide the problem to
subtasks and then to measure the goodness of the resulting align-
ments. User-friendly tools to use PRANK for phylogenetic analyses
will be provided shortly.
Although an iterative search strategy should help PRANK to
greatly reduce the problems caused by an incorrect start guide
phylogeny, iteration does not decrease the greediness of the algo-
rithm nor can it solve the phylogeny for datasets that have no
unique phylogeny, e.g., due to incomplete lineage sorting. Both
challenges can be tackled by re-implementing the phylogeny-aware
algorithm for the alignment sequence graphs [ 16 ]. By using addi-
tional edges to indicate unresolved gaps and then pruning the
unused edges after the alignment of the next sequences, one can
implement an algorithm very similar to that of PRANK +F (Fig. 4 ).
The advantages of the graph approach are greater, though, if the
edges are not pruned but given weights or probabilities based on
the evidence for the different mutation types. Such edge-weighting
approach makes the method far less sensitive to errors in the guide
phylogeny or different sites evolving under slightly different phy-
logenies. We are actively implementing the features of PRANK still
missing from the graph approach and believe that the new method,
called PAGAN , will soon replace PRANK .
As discussed above and evidenced by methods for co- and joint-
estimation of alignment and phylogeny, the multiple sequence
alignment should always be seen with the associated phylogeny.
Understanding the alignment and drawing right conclusions from
it is much easier when the relationships between the sequences are
indicated next to the alignment. For a method such as PRANK ,
the phylogeny is also needed to indicate the relative positions
of the ancestral sequences and to visualize the changes happening
in different evolutionary branches. We are developing Wasabi, a
browser-based graphical user interface, that integrates these ideas