Classifying Conserved RNA Secondary Structures With Pseudoknots by Vector-Edit Distance

Secondary structures have been proved to relate with the great functional diversity of RNA. There have been many studies to predict and compare the RNA secondary structures. However, fast and accurate comparison of RNA secondary structures with arbitrary pseudoknots is a challenging issue due to the hidden but important structural properties, such as the distribution of stems and branches. In this paper, we construct a novel RNA secondary structure model called modified adjoining grammars binary tree (
$\mathrm {BTMG_{CSP}}$
). It can not only represent the complex RNA secondary structure including arbitrary pseudoknots intuitively, but also reserve RNA secondary structure properties. Further, we propose vector-edit distance to measure the structure similarity between
$\mathrm {BTMG_{CSP}}$
trees converted from RNA sequences and their secondary structures for classifying conserved stem pattern. The experimental results show that our method substantially reduces the memory and time consumption in contrast to previous algorithms, such as
$O((n/k)^{2}) $
and
$O(n/k) $
for time and space complexity, respectively. In particular, the AUC value of our method achieves 0.949 in PseudoBase.