Phylogeny-aware alignment with PRANK - Multiple Sequence Alignment Methods - page 158

Some of the limitations of the approach and the measures to
overcome them can be contradicting. Accurate calling of insertion
and deletion events requires densely-sampled sequence sets but the
inference of alignment phylogeny for a large dataset is prone to
errors [ 9 ] and the alignment may therefore suffer. Furthermore,
incomplete lineage sorting is more likely among closely-related
sequences and possibly no single phylogeny correctly reflects the
evolutionary history of all sites of a very densely-sampled sequence
set. As discussed below, some of these contradictions can be over-
come with a more advanced implementation of the algorithm.

5 Practical Advice for the Use of PRANK

Evolutionary sequence analysis is based entirely on multiple
sequence alignment and the accuracy of the downstream analysis
depends on the correctness of the underlying alignment. Align-
ments produced with PRANK have been shown to provide exception-
ally accurate inferences of selection on protein-coding sequences
[ 10 , 11 ] and perform well in phylogenetic analyses [ 12 ], although
the latter finding is somewhat controversial due to the role of the
guide phylogeny in the phylogeny-aware alignment. Despite its
good performance in evolutionary analyses, PRANK is sensitive to
violations of the assumptions made by the algorithm and the users
of the program should understand the requirements and limitations
of the method.

Alignment phylogeny: The phylogeny-aware alignment algorithm
uses the alignment guide phylogeny to distinguish insertions from
deletions. The algorithm is therefore sensitive to errors in the guide
phylogeny, the variant with permanent insertions ( PRANK +F ) being
especially sensitive. Any PRANK alignment should be performed
using an accurate guide phylogeny: if a high-quality phylogeny is
available for the sequence set, it should be used instead of the
heuristic phylogeny inferred by the program. (See below for future
plans on integrated phylogeny search with PRANK .)

Evolutionary distances: PRANK uses the branch lengths provided by
the guide phylogeny to re-compute the substitution and gap scor-
ing for each alignment step. Depending on the expected evolution-
ary divergence, a region with several dissimilarities may be
considered homologous and matched (distant sequences), or
non-homologous and placed in separate columns (close
sequences). Although the algorithm is not sensitive to small devia-
tions in the branch lengths provided, a guide phylogeny with
accurate distance estimates should be used when available.

Option +F: Given that the alignment guide phylogeny is correct
and the sequence sampling is dense, the variant with permanent
insertions ( PRANK +F ) has been shown to outperform the basic