RNA secondary structure is predicted by computing the structure with the
minimum free energy. Although RNA structure without pseudoknots can
be found using dynamic programming algorithms, finding general
structures with pseudoknots is a nondeterministic polynomial-time (NP)
hard problem. Several methods, such as recursive simple pseudoknots, have
been developed in the past for obtaining a conformation with globally
minimal energy among the restricted class of pseudoknots. In this work, we
develop a new method for approximating a conformation with low energy,
posing no restrictions on type of pseudoknots contained in the RNA
secondary structure. In our method, the low-energy RNA secondary
structure is obtained by repeatedly removing helices and performing
dynamic programming to obtain the structure with energy lower than that
obtained in the previous iteration. This method can be considered as a local
minimization, and can be combined with any global optimization method
that takes advantage of local minimization. We tested performance and
convergency of the method by predicting a secondary structure of several
RNA sequences, which is a priori known to contain pseudoknots.