Figure 4

For most alignment algorithms, sequence duplications can lead to serious mis-alignments and to instabilities under minor modifications of the input sequence sets. For our original sequence set = {S 1, ..., S 22} and the 'reduced' sequence sets _i := \{Si}, we obtained six classes of multi-alignments. Here, class 3 and 5 seem to be the only types of alignments that are biologically meaningful. Alignments of class 5 were obtained only once, namely by DIALIGN using the xfr option as described in 2 and 4.3. We used this option to prevent the program from aligning the motifs A₂, S₂ and B₂ in the sequences S20 – S22 to the motifs A, S and B in the remaining sequences.