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

Masaki Tagashira

The PhyloFold program performed the best trade-off of the above metrices (Figure 3 ) among the PhyloFold, CentroidFold (Hamada et al., 2009b) , CentroidHomFold (Hamada et al., 2009c) , and TurboFold-smp (Tan et al., 2017) programs (Table 2 ) on test set "unaligned" (Table 3) while demanding comparable running time (Table 4 ). Ten percent of ncRNA families whose reference seed structural alignments have at most 200 columns and whose number of homologs is less than 11 are sampled from the Rfam database (Kalvari et al., 2018) . The reference seed structural alignments of the 147 sampled families are compiled as test set "aligned". The rightmost column shows the type of probabilities used to predict secondary structures. Precise alignment type is proposed in this study. Decomposed alignment type considers structural alignment to decompose an alignment STA RNA into the structures and alignment SS RN A , SS RN A , SA RNA . Single sequence type does not take support homologs into account. "+ consensus structure" means that probabilities on consensus secondary structure given input sequence alignment are mixed. Precise alignment (+ consensus structure) type is expected to display the highest prediction accuracy. Unaligned homologs of 11 ncRNA families and the reference secondary structures of the homologs were collected from the RNA STRAND database (Andronescu et al., 2008) and compiled as test set "unaligned". The Rfam database was not used to build the set because this database does not register reference single secondary structures. Ten homologs and their reference secondary structures were sampled from each ncRNA family when the family had 11 or more homologs. The middle and rightmost columns show the number of homologs and the range of lengths on each ncRNA family, respectively. "Coronavirus 3 prime stem-loop…" and "Vimentin 3 prime UTR" stand for "Coronavirus 3 prime stem-loop II-like motif (s2m)" and "Vimentin 3 prime UTR protein-binding region," respectively.