Nearest neighbor thermodynamic approaches are the cornerstone of popular RNA secondary structure design and prediction tools. These include Mfold,1 the ViennaRNA package,2 the RNAstructure package,3 Sfold,4 NUPACK,5 and the LinearX6, 7 tools. The functional form is also used for programs that learn parameter values from known structures,8 including RNAsoft,9 TORNADO,10 ContraFold,11 and Eternafold.12

Nearest neighbor approaches estimate folding stability of a given secondary structure compared to a random coil as the Gibbs free energy change. The approach is called nearest neighbor because the free energy contribution of each motif, i.e., Watson-Crick-Franklin helices, bulge loops, internal loops, hairpin loops, multibranch loops, and exterior loops, depends on the sequence in the motif and the sequence of adjacent base pairs. The free energy of a given secondary structure is estimated by summing the free energy increment of individual components of the secondary structure.13, 14, 15, 16, 17

A nearest neighbor model consists of two components: the functional form that embodies empirical rules and the parameter values used in those equations. The parameter sets provided by the NNDB were derived from optical melting experiments on small model systems. These parameter sets are used in many software packages for structure prediction.