Discussion
The algorithm presented here does not take into account the effect of relative GC content and stacking interactions of neighboring bases on the melting temperature of the oligo-nucleotides. Accordingly, the oligo-nucleotides suggested by the program can differ significantly in melting temperature. However, as this can easily be adjusted after the selection is made, we have not included a subroutine that takes GC content into account during the primary search, because this would slow down the calculations. Furthermore, we expect that GC content differences may be of less importance for the applications envisioned here, because they can be largely compensated by the choice of experimental conditions, such as buffers that compensate stability differences [13].
A more general problem is our way of calculating the relative stability factor. This does currently not take the nucleotide composition into account either. The reason is that there are too few experimental data as yet, that would allow to unequivocally include this in the calculations. The current experimental data sets focus on the types of mismatches in particular contexts, but not systematically on position specific effects [7,15]. Moreover, they deal with relatively short model oligos only (up to 12 nt). However, the probes used for species identification are longer and the different effects can currently not be accurately assessed from experimental data for such longer probes. In our equation, it is mainly the border parameter n that would be affected by base composition and nearest neighbor interactions and we have therefore left this as a variable that can be set according to experimental results. In principle, it seems possible that n differs for different sequence compositions, i.e. GC-rich stretches have a smaller n than AT-rich ones. Thus, if one chooses a low n, one would risk that GC-rich oligos are suggested as specific probes that still show cross hybridization. However, it seems that these can easily be eliminated after the selection is made. Still, if experimental data indicate that this is a major problem, the program could easily accommodate such new insights.
Finally, the stability function proposed in Equation 1 could possibly also have other shapes than Gaussian. Again this is a factor that needs further experiments. If it turns out that other functions are more appropriate, one can include this as additional options into the program. At the present we offer the extreme, namely a flat function, as an alternative option.