"... In PAGE 5: ... Corona_pk3 and HDV_ribozyme constitute simple pseudoknot structures which can be analyzed by an SL- TAG,whereasTombus_3_IVhasonebranchingsecondarystructure involving a pseudoknot which requires an ESL-TAG. The results in Table2 show that PSTAG can predict accurate structural alignments for all three RNA families.... ..."

"... In PAGE 2: ... Therefore, it is necessary to take into account both structural and sequential information in comparing RNA sequences. Several tools are available that carry out RNA alignment and folding at the same time ( Table1 ). The pioneer work by Sankoff [17] involves simultaneous folding and align- ing of two RNA sequences, and has huge time and space complexity (Table 1).... In PAGE 2: ... Several tools are available that carry out RNA alignment and folding at the same time (Table 1). The pioneer work by Sankoff [17] involves simultaneous folding and align- ing of two RNA sequences, and has huge time and space complexity ( Table1 ). FOLDALIGN [18] improves the Sankoff apos;s method by (1) scoring the structure solely based on the number of base pairs, instead of the stacking ener- gies; and (2) disallowing branch structures (junctions).... In PAGE 2: ... [22] presented RAGA to conduct alignment of two homologous RNA sequences when the secondary structure of one of them was known. As shown in Table1 , most of these methods suffer from high time complexities, making the structure-based RNA alignment tools much less efficient than sequence-based alignment tools. Tools that search for optimal alignment for given struc- tures include RNAdistance [25], rna_align [26], and RNA- forester [27].... In PAGE 2: ... With the crossing arcs, rna_align is able to align two RNA secondary structures, one of which could con- tain pseudoknots. RNAforester extends the tree model to forest model, which significantly improves both time and space complexities ( Table1 ). In addition, methods using Stochastic Context Free Grammars (SCFGs) have been developed to compare two RNA structures.... In PAGE 2: ... stemloc uses quot;fold envelope quot; to improve efficiency by confining the search space involved in calculations. The time and space complexities of these two tools are also listed in Table1 . Furthermore, pattern- based techniques such as RNAmotif, RNAmot and PatS- earch [3,33,34] have been used in database searches to detect similar RNA substructures.... ..."

"... In PAGE 8: ... In order to evaluate the impact of the parameter k on the accuracy of the algorithm, we carried out the same searching experi- ments for each given k. Table1 shows the number of filters selected for each tested structure and the fil- tration ratio for the one that is first applied to scan the genome. Table 2 shows that on the tested RNA families, the filtration based approach achieves the same or better searching accuracy than that of the original approach.... ..."

"... In PAGE 8: ... In particular, a significant improvement on specificity is observed on a few tested families. From Table3 , compared to the original approach, the filtration based approach consumes a significantly re- duced amount of computation time. On most of the tested families, the filtration based searching is more than 30.... ..."

"... In PAGE 8: ... It is further interesting to note that a substantial number of potential donor atoms and an even larger number of acceptor atoms is not involved in H-bonds. In Table2 the H-bonding residue interaction table is shown. There is only one nucleotide not involved in any H-bond: C21.... In PAGE 8: ...able is essentially identical to the structure given by Shen and Tinoco (1995) in their Fig. 9. There is, however, one borderline case. Table2 shows that there is a H-bond between U34 and U8, the latter forming a Watson-Crick pair with A33. According to our definition of secondary structure this H-bond is part of a helix-like secondary structure element.... In PAGE 9: ... In this case, however, the H-bonds G4:O2*-C5:O4* and G4:O2*-A26:N1 are alternative. Further, there are various other nucleotides directly bonded to four, three or two nucleotides, see Table2 . Of course, these poly-nucleotide interactions include weak H-bonds.... ..."

"... In PAGE 12: ... We consider a real hit as comprised of hits that are from the results for different pieces and contiguous in locations on the genome. Table9 shows the results of our experiments. Our approach successfully identified a complex multiple pseudoknot structure in viral genomes at essentially the correct location; however, some portions of these complex structures were not found correctly.... In PAGE 12: ... Our approach successfully identified a complex multiple pseudoknot structure in viral genomes at essentially the correct location; however, some portions of these complex structures were not found correctly. It can be seen from Table9 that the searching algorithm is able to recognise most of the structural signals of the pseudoknot structures in this particular domain. Pk1 is not found on genomes of TMVC, TVV and RV at the corresponding locations where it is present in those of TMVF and TMV.... ..."

"... In PAGE 4: ...RNA Number of training sequences Number of nucleotides Pseudocount tmRNAa49 pk12 a50a25a51 a52a53a50a8a54a55a49a57a56a25a58a31a54 a52a25a59 a58 tmRNAa49 pk34 a60a25a61 a61a25a54a62a49a63a52a64a56a31a54 a56a65a59 a66 srpRNA a56a31a66 a50a25a54a67a49a57a58a8a54 a52a25a59 a56 telomerasea49 vert a52a68a50 a61a8a54a55a49a57a56a8a54a8a54 a54a14a59 a61 coronaa49 pk3 a52a68a66 a51a25a54a67a49a57a69a8a54 a54a14a59 a61 HDVa49 ribozyme a52a53a58 a61a8a54a55a49a63a52a53a54a8a54 a52a25a59 a54 tombusa49 3a49 IV a52a64a69 a61a25a54a62a49a63a52a53a54a8a54 a52a25a59 a54 alphaa49 RBS a61 a52a53a54a8a54a55a49a63a52a64a56a31a54 a54a14a59 a60 antizymea49 FSE a52a68a50 a58a31a54a67a49a57a51a8a54 a54a14a59 a61 IFNa49 gamma a58 a52a53a51a8a54a62a49a70a52a68a60a25a54 a54a14a59 a51 determined with methods similar to the ones introduced by Klein and Eddy [11] before the search starts. As can be seen in Table2 , the program correctly identifies more than a38a9a7a39a15 of in- serted sequence segments with excellent specificity in most of the experiments. The only exception is the srpRNA, where the program misses more than a34a14a7a39a15 inserted se- quence segments in one of the experiments.... In PAGE 4: ... Additionally, while the alphaa21 RBS pseudoknot has a more complex structure and three CM components are needed to model it, our searching algorithm efficiently identifies more than a36a39a34a9a15 of the inserted pseudoknots with high specificities. A higher Ca43 G concentration in the background does not adversely affect the specificity of the model; it is evident from Table2 that the program achieves better overall performance in both sensitivity and specificity in a background of higher Ca43 G concentrations. We therefore conjecture that the specificity of the model is partly deter- mined by the base composition of the genome and is improved if the base composition of the target gene is considerably different from its background.... ..."