The methods above the dashed line are purely sequence-based alignment methods. RNA structural alignment methods are listed below the dashed line. The names of MAFFT options are shaded. The benchmark dataset in the MASTR paper [87] was used. The alignment accuracies were assessed with two criteria, SPS and SCI [110, 111], using the compalign and scif programs distributed with the BRAliBASE Version 2.1 benchmark dataset [29]. The SPS SCI values were computed for each alignment and then averaged across all the alignments. The alignment by each method was subjected to three external prediction programs, Pfold [112], McCaskill-MEA [90] and RNAalifold [89], and then the differences from the Rfam curated structure were calculated with Matthews correlation coefficient (MCC) criterion. The accuracy values for secondary structure internally predicted by RNA Sampler and MASTR are shown in the (intrinsic) column. The MCC values were computed for each sequence and then averaged across all the sequences. The highest accuracy values are underlined for each column. The accuracy values close to the highest (P > 0.01 in the Wilcoxon test) are shown in bold. McCaskill-MEA was run with the default α value of 0.91. RNA Sampler was run with the -i 15 -S 1∅∅ arguments. See Katoh and Toh (submitted for publication) for benchmark results using larger datasets. See the MASTR paper [87] for the results of other methods, including FoldalignM, LocARNA and RNAcast, that are not listed here.