Fig. 1

Comparison of structure-based benchmarks commonly used to rank sequence alignment programs. Each contour line surrounds the densest 75 % of points representing pairwise alignments in the benchmark. Structural identity is based on matching 8-state DSSP [34] secondary structure assignments (see Methods). Perfect secondary structure agreement would result in a score of 1 on the y-axis. Reference alignments exhibit decreased structural similarity as the distance between sequence pairs increases. Realignments using MUSTANG [40] showed improved quality in some cases, especially relative to the original SABmark [43] reference alignment