Choice of substitution matrices Each alignment program has to use a substitution matrix for replacement of characters during the alignment process. Traditionally these matrices differentiate between transitions (purine to purine and pyrimidine to pyrimidine substitutions) and transversions (purine to pyrimidine and vice versa), but more complex matrices have been described in the literature. An example for the latter are the RIBOSUM matrices [30] used by RSEARCH to score alignments of single-stranded regions. To address the question whether incorporating RIBOSUM matrices results in a significant performance change, we used the RIBOSUM 85–60 4 × 4 matrix as substitution matrix for CLUSTALW, ALIGN-M and POA, as these programs allow an easy integration of non-default substitution matrices via command line options. Since gap-costs and substitution matrix values are interdependent we adjusted the original RIBOSUM values to the range of the default values. We applied Wilcoxon tests to test whether using the RIBOSUM matrix (instead of the simpler default matrices) yields a statistical significant performance change. Results are summarized in Table 4. POA and ALIGN-M perform significantly better, only CLUSTALW's performance suffers from RIBOSUM utilization. The reason for CLUSTALW's performance loss is not obvious to us; it might be that CLUSTALW's dynamic variation of gap penalties in a position and residue specific manner [27] works optimally only with CLUSTALW's default matrix. Furthermore, the RIBOSUM 4 × 4 matrix is based on nucleotide substitutions in single-stranded regions whereas we used it as a general substitution matrix. Other matrices, based on base-paired as well as loop regions from a high-quality alignment of ribosomal RNA [31], gave, however, no significantly different results (data not shown). Table 4 Comparison of default vs. RIBOSUM substitution matrix by Wilcoxon tests Program k2 k3 k5 k7 k10 k15 ALIGN-M / + + + / / CLUSTALW - - - - - - POA + + + / / / If the use of the RIBOSUM 85–60 matrix resulted in a statistically significant performance increase in comparison to use of the default matrix this is indicated with a "+"; "-" indicates that the default matrix scores significantly better. If no statistical significance was found this is indicated with a "/".