This implies, that improved optimisation algorithms will not lead to biologically improved alignments for these sequences. In this case, however, there is some correspondence between sequence similarity and homology, so one should hope that the performance of DIALIGN on these data can be improved by to designing better objective functions. An interesting example from BAliBASE is shown in Figure 4. Here, the non-anchored default version of our program produced a complete mis-alignment. However, it was sufficient to enforce the correct alignment of one single column using corresponding anchor points to obtain a meaningful alignment of the entire sequences where not only the one anchored column but most of the three core blocks are correctly aligned. This indicates that the correct alignment of the core blocks corresponds to a local maximum in the alignment landscape.