Through the RF-predict tool, rPredictorDB supports Minimum Free Energy secondary structure prediction using the RNAfold algorithm of Hofacker et al., implemented in the Vienna RNA package. RNAfold is one of the most famous and widely used secondary structure prediction algorithms and is the leading implementation of the Free Energy Minimalization principle.
The algorithm searches for a chemically most stable secondary structure, where “stability” means minimizing the free energy of the secondary structure. The assumption is that the molecule will in “real life” tend to have the structure that is most stable. In this approach, a large thermodynamic model (over 7600 parameters) is used to describe the thermodynamic stability of various small structural features: base pairs, small hairpins, bulges, etc. Then, a dynamic algorithm of Zuker and Stiegler (1981) is used to compute the set of base pairs that optimizes the free energy of the structure according to the thermodynamic model.
There are two shortcomings of this approach. First, the thermodynamic model does not describe larger-scale structures well. Second, because of computational (in)efficiency, the dynamic programming algorithm is restricted to structures without pseudoknots. Ribosomal RNA, unfortunately, are both long and pseudoknotted. The sub-par results RNAfold obtains on rRNA structures and other properties of rRNA like good evolutionary conservation have prompted us to develop a custom secondary structure prediction for rRNA: CP-predict: a two-phase algorithm for rRNA structure prediction.
The original RNAfold paper is available at the University of Vienna website (Note that the file is in PostScript format.)