Relative performance of RNA sequence alignment programs
To find the sequence alignment program that performs best under all non-trivial situations (e. g. reference APSI ≤ 80 %), we did a comparison of all those programs previously identified [22] to be top ranking. If available we used the newest program versions and optimized parameters. In the comparison we included the RNA version of PROBCONS [33] (PROBCONSRNA; see [34]) whose parameters have been estimated via training on the BRAliBase II datasets. We applied Friedman rank sum tests to each alignment set with a fixed number of sequences. Results are summarized in Table 5. MAFFT version 5 [25] with the option "G-INS-i" ranks first throughout all test-sets. This option is suitable for sequences of similar lengths, recommended for up to 200 sequences, and uses an iterative (COFFEE-like [35]) refinement method incorporating global pairwise alignment information. This option clearly outperforms the default option "FFT-NS-2", which uses only a progressive method for alignment. MUSCLE and PROBCONSRNA rank second and third place.
Table 5  Ranks determined by Friedman rank sum tests for all top-ranking programs.
Program/Option  k2  k3  k5  k7  k10  k15
CLUSTALW (default)  8  7  8  8  7  7
CLUSTALW (optimized)  6  6  7  7  6  6
MAFFT (FFT-NS-2)  2  4  4  4  5  5
MAFFT (G-INS-i)  1  1  1  1  1  1
MUSCLE  3  3  3  2  2  2
PCMA  9  10  10  10  10  10
POA  7  8  9  9  9  9
PROALIGN  5  5  6  6  8  8
PROBCONSRNA  4  2  2  3  3  4
PRRN  10  9  5  5  4  3
Programs were ranked according to BRALISCORE averaged over all alignment sets with k ∈ {2, 3, 5, 7, 10, 15} sequences and APSI ≤ 80 %. MAFFT (G-INS-i) is the top performing program on all test sets. For program versions and options see Methods.