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

Sperschneider, Jana; Datta, Amitava

One major advantage of the detection approach is that the pseudoknot prediction target class is predefined and transparent. In contrast, dynamic programming algorithms construct pseudoknots through their recursion scheme, which in some cases leads to unspecific pseudoknot target classes (24, 57) . In this work, recursive H-type pseudoknots will be considered similar to the class of pseudoknots predicted by pknotsRG (26) . A recursive H-type pseudoknot has two crossing stems S 1 and S 2 , resulting in three loops L 1 , L 2 and L 3 . All of the three loops are allowed to form internal secondary structures; however, loop-loop interactions are not allowed. In this work, we also include pseudoknots where one of the stems S 1 or S 2 is interrupted by bulges or internal loops. This leads to a more comprehensive prediction class as in pknotsRG, where only bulges of size 1 nt are considered (26) . In general, the three dictionaries D s , D L s and D M s allow for the construction of complicated pseudoknot folds. For example, it is straightforward to construct kissing hairpins from the stem dictionary. The pseudoknot energy parameters become the bottleneck in this approach, not necessarily the complexity of pseudoknots. The three main steps for constructing recursive H-type pseudoknots are shown in Figure 4 . First, so-called core H-type pseudoknots form through simple combination of two crossing stems. They become recursive H-type pseudoknots when additional secondary structure elements fold in each of the three loops. Note that recursive pseudoknots are not allowed in the loops as this may lead to sterically infeasible configurations. The overall recursive H-type candidate pseudoknot is assembled in a third step.