PhyloFold: Precise and Swift Prediction of RNA Secondary Structures to Incorporate Phylogeny among Homologs

Masaki Tagashira

The components f e SS RN A , f e SS RN A can be computed by the estimated parameters of the Turner (nearest neighbor) model, which approximates the free energy of RNA secondary structure on thermodynamics. (Turner and Mathews, 2010) As the components, conventional methods based on structural alignment employ posterior pairing probability matrices on single secondary structure, predicted by inside-outside algorithms such as the McCaskill algorithm (McCaskill, 1990) , to simplify computations, although the suitability of the matrices has not been discussed. (Hofacker et al., 2004; Do et al., 2008) . Hence, the Turner model, instead of the matrices, is adopted in this study to prevent a reduction in the accuracy of a matrix P PA RNA . The components f e PA , f e LA can be computed by the RIBOSUM score matrices, predicted from learning datasets of validated structural alignments. (Klein and Eddy, 2003) In this study, the RIBOSUM score 80-65 matrices, which are the most popular, are adopted.