Fig. 2
From: Scaling statistical multiple sequence alignment to large datasets
Results on 10,000 sequences. Using UPP on two different backbones: one computed using default PASTA and the other computed using PASTA+BAliPhy (i.e., one iteration of PASTA using BAli-Phy as the subset aligner after default PASTA completes). Each point represents one replicate. Delta-RF refers to the difference between the RF error rates of ML trees computed on the estimated and true alignments. In each subfigure, a position above the 45-degree line indicates that PASTA+BAli-Phy is preferable; the axes for the subfigure for Delta-RF have been flipped to maintain this interpretation, since Delta-RF is an error metric rather than an accuracy metric