Fast Algorithms for RNA Secondary Structure Prediction

RNA secondary structure prediction with pseudoknots is important, since pseudoknots are part of functionally improtant RNAs in cells. State of the art dynamic programming algorithms due to Akutsu et al [7] and Deogun et al [8] perform well on single RNA sequences. Our aim of this project is to be able to predict secondary structure of real life RNA sequences, which can be more than 700 nucleotides length. Recurrence for solving simple pseudoknots involve O(n) time and O(n) complexity. Existing dynamic programming algorithms solve this recurrence over full length of RNA sequence, which make them impractical for sequences of length greater than 500 nucleotides. We propose novel two-stage dynamic programming algorithm which can overcome this drawback by predicting regions of pseudoknots in first stage and predicting actual structure of pseudoknots in second stage. Another scalable approach HxMatch, by Witwer et al [15] predicts secondary structure of consensus sequence of multiple aligned RNA sequences. HxMatch uses Maximum Weighted Matching (MWM) [30], which has O(n) time complexity, for choosing basepairs based on score-matrix. Few observations suggest that score-matrix is so good that even simpler algorithms can be used for choosing basepairs rather than MWM. We propose to use Random Sampling algorithm with time complexity of O(n logn) for choosing basepairs, instead of MWM. We performed experiments on Pseudobase dataset and other real life dataset taken from Rfam and NCBI. Twostage algorithm outperforms both Deogun’s algorithm and Rivas’ algorithm in terms of time, however, prediction accuracy is less than Deogun’s algorithm but better than Rivas’ algorithm. Random Sampling algorithm performs as good as MWM for most of the sequences.