A comprehensive alignment-filtering methodology improves phylogeny particularly by filtering overly divergent segments

Abstract

Abstract​Alignment problems may be complicated and diverse and the performance of the existing alignment-filtering tools, particularly their potential effect on phylogenetic reconstruction, have remained debated or limitedly explored.​In the present study, we developed a new R package named alignmentFilter to treat the diverse alignment problems, especially masking ambiguously-aligned or overly divergent segments using a newly devised grouping-regrouping algorithms acting on sequence divergence in each of sliding windows throughout alignment. Then we tested and compared the power and accuracy of the prevalent alignment-filtering tools, particularly the effect on phylogeny based on Angiospermae genome-scale data and simulated data.​The results indicate the alignment-filtering methods alone may affect the phylogeny decisively, producing (strongly-supported) phylogenetic conflicts. In most cases, alignmentFilter most efficiently improves phylogeny, yielding more congruent phylogeny regardless of phylogeny-reconstruction methods, producing phylogeny more concordant to the priori tree regarding to both topology and branch length in simulation, and most efficiently decreasing the root-to-tip length heterogeneity that is implied to be negatively correlated with topological accuracy.​The concepts and algorithms of alignmentFilter are quite distinct from existing alignment-filtering methods. It is not susceptible to specific types of alignment errors that usually baffle other methods and can reach the optimal solution frequently with lowest computation complexity. The biological or statistical sense of the key optional stringency-controlling parameter is more straightforward and the setting and outcome may be more adjustable or predictable. A better and more comprehensive solution of alignment problems in this study will benefit tremendous downstream analyses sourced from alignment and phylogeny.