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

Sperschneider, Jana; Datta, Amitava

We evaluated our algorithm on a set of pseudoknotted and pseudoknot-free sequences of different RNA types ( Table 2) . Given a sequence, DotKnot is a method that predicts only pseudoknots. Therefore, predictive accuracy is measured for base pairs belonging to a pseudoknot. Two measurements are used for comparison of DotKnot and other selected algorithms from the literature. For each published pseudoknot in a sequence we report sensitivity S ¼ 100 Â TP=TP þ FN ð Þ and the positive predictive value PPV ¼ 100 Â TP=TP þ FP ð Þ . True positive (TP) corresponds to the number of correctly predicted base pairs in the predicted pseudoknot, False negative (FN) to the number of base pairs in the published pseudoknot that were not predicted and False positive (FP) to the number of incorrectly predicted base pairs in the predicted pseudoknot. A pseudoknot is said to be predicted by an algorithm if it is a crossing structure element and at least one of the two pseudoknot stems is partially predicted. Furthermore, the ratio r ¼ (number of correctly predicted pseudoknots)/(number of predicted pseudoknots) is reported. We compare DotKnot to two dynamic programming methods, namely pknots (24) and pknotsRG (26) , the pseudoknot detection tool KnotSeeker (39) and the heuristic approach HotKnots (35) . HotKnots returns a number of sub-optimal scenarios; however, we only evaluate predictive accuracy for the best solution.