Values of sensitivity and positive predictive value (ppv) for RNAfold and RNAenn with respect to various RNA families.

Sensitivity is the ratio of number of correctly predicted base pairs divided by the number of base pairs in the native structure; positive predictive value is the ratio of the number of correctly predicted base pairs divided by the number of base pairs in the predicted structure. Since RNAenn currently does not include energy contributions for dangles (single stranded, stacked nucleotides), RNAfold was used without dangles (version 1.8.5 with -d flag). To our knowledge, there has not been a careful benchmarking of structure prediction accuracy between the Turner 1999 energy model and the newer Turner 2004 energy model, though it is interesting to note that RNAenn has better structure prediction when using Turner 1999 for base stacking. Overall, it is clear that RNAfold outperforms RNAenn (Turner99), although a few cases, such as ec and rnap2 RNAenn have better sensitivity. Nevertheless, we expect much better performance in the future when our triplet and base stacking energy terms have been refined by using knowledge-base potentials. The database of RNA structures in this benchmarking set comes from a data collection of D.H. Mathews (personal communication), which derives from published databases [26], [54], etc. See [55] for a citation of original data sources.