311
considered. Not every folding is thermodynamically optimal (it should always have a low
folding energy, because the lower the free energy, the more stable the structure), especially
since there are several secondary structure forms (e.g. stem-, hairpin- and interior-loop).
Secondary structures can be predicted bioinformatically, but this is not easy. There are
various algorithms for this purpose, which are all based on dynamic programming meth
ods, but nevertheless work differently. For example, the Nussinov algorithm first calcu
lates the maximum number of base pairs and then uses this information to calculate the
secondary structure with the maximum base pairing. However, since RNA structures do
not always have the maximum possible base pairing, this method does not always give
useful results. A more optimal and faster solution for structure determination is provided
by algorithms based on energy minimization. The Zuker algorithm calculates the optimal
secondary structure with the minimum free energy, based on a thermodynamic model, e.g.
mFold server. On the other hand, the Sankoff algorithm simultaneously folds and aligns
two sequences using an energy model to minimize the free energy, e.g. LocARNA pro
gram. A useful online web server for secondary structure prediction is ViennaRNA
Webservices (https://rna.tbi.univie.ac.at/). There are many more tools for RNA analysis
here. For additional information, see the book section or Kunz et al. (2015).
In the exercise example, RNAfold (also in ViennaRNA Webservices, also based on
energy minimization) should find a possible secondary structure fold with a minimum free
energy of − 360.20 kcal/mol.
20.2
2. Here it is important to see that the change in energy released is not automatically equal
to the sequence length, e.g. it is not double. For example, the sequence
ATGCTACGCGATGCATCGAGCGCAT has an energy of −3.5 kcal/mol and twice the
20.2 Magic RNA