DotKnot: pseudoknot prediction using the probability dot plot under a refined energy model

Sperschneider, Jana; Datta, Amitava

It has been shown that choosing base pairs with high probability can improve secondary structure prediction (50) . There are also several approaches that discuss the exclusion of base pairs with low probability for improving runtime. Here, the common technique is to use a cut-off value for base pair probabilities in order to determine only significant base pairs (51) . However, for pseudoknot stems we cannot simply dismiss base pairs with low probability. The partition function calculation is based on the ensemble of secondary structure elements, which are non-crossing interactions. In general, the pseudoknot stems are visible in the probability dot plot as they are members of the folding space. One can expect that at least one of the pseudoknot stems will have low base pair and stack probabilities. By default, RNAfold only displays probabilities larger than 1 Â E À5 . For pseudoknot detection in the dot plot, we use a cutoff probability of 1 Â E À11 to make sure that all pseudoknot stems can be found for long sequences. Given the probability dot plot, stems are assembled according to certain criteria. First, a stem must have at least three base pairs. Second, one can expect that the stack probabilities in a stable stem do not rise or drop sharply (49) . Helix elongation is an energetically favourable process; however, wobble base pairs can be destabilizing (8, 52, 53) . Furthermore, there may be other stable secondary structure elements competing for base pairs. Therefore, we demand that the absolute percentage increase or decrease of stack probabilities for subsequent base pairs ði, jÞ, ði þ 1, j À 1Þ in a stem has to be smaller than a certain threshold .