Abstract
Problems in direct optimization method as implemented in the software POY were reviewed. Alignment of highly variable regions using POY is problematic because it likely provides artificial maximization of pseudo-homology within the alignment. If the variable regions are aligned with conserved regions, then the bias could be emerged in two ways. When the conserved regions contain only weak phylogenetic signal, then stochastic similarities within the variable regions strongly affect to the final alignment. This type of bias can be emerged even if the variable regions are aligned independently. As a result, aligned variable sequences may provide superior amount of pseudo-signal which could even mask true weak signal contained in the conserved regions. When the conserved regions contain significant phylogenetic signal, then the variable regions will play as adherents of the conserved data even if the variable regions contain no or even contradicting phylogenetic signal. The adherent bias is especially problematic when data set contains heterogeneous sequences by gene introgression, recombination, or lineage sorting. Structure- or similarity-based approaches should be used in aligning DNA sequences.
