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

Sperschneider, Jana; Datta, Amitava

The lack of loop entropy parameters is a critical issue in pseudoknot prediction (43) . For H-type pseudoknots with interhelix loop size 1 nt, loop entropy values were derived using several fitted parameters (41) . Gaussian chain approximation based on polymer physics for pseudoknot loop entropies was also proposed (44) . Lattice-based models were developed to take into account volume exclusion effects (45) (46) (47) . The most successful models are based on the atomic coordinates of the RNA backbone. Using polymer statistical mechanics, Cao and Chen (42) calculated loop entropy parameters for pseudoknots with interhelix loop size 1 nt. Recently, their so-called virtual bond model has been extended to pseudoknots with interhelix loop size 6 nt (48) . Inclusion of such a pseudoknot energy model is not straightforward in dynamic programming in reasonable runtime due to dependencies between opposite loops and stems. Calculation of the full partition function under the virtual bond model takes Oðn 6 Þ time and Oðn 2 Þ space, making the approach only feasible for sequences shorter than say 150 nt (42) . However, pseudoknot energy parameters derived from the virtual bond model are readily available in tabular form for many stem and loop length combinations (42, 48) . It is straightforward to incorporate such energy parameters in a pseudoknot detection approach, making prediction using much improved energy models feasible for long sequences.