A CRF Based Algorithm For RNA Secondary Structure Prediction

When more and more non-coding genes and their functions have been identified and revealed, the researchers come to realize the importance of non-coding RNAs. Because the structure determines the function, it is very important for the studying of RNA secondary structures. Because of the large number of non-coding RNAs and the limitations of biological experiments, RNA secondary structure prediction becomes a important way for researching the function of non-coding RNAs.There are two major directions for RNA sequence research methods. They are the RNA secondary structure prediction methods based on multiple comparative sequence analysis and RNA secondary structure prediction methods based on a single sequence. This paper focuses on the single-sequence secondary structure methods. The base pair maximization algorithm and minimum free energy algorithm are used to deal with the prediction of RNA secondary structure, and they are deterministic dynamic programming algorithm, but they are not good solution result from the accuracy issues. In this paper, we also introduced Hidden Markov model (HMM) and random context-free grammar (SCFG), and they are based on probability theory, but these two algorithms has a problem of inductive bias data as well as the problem of the higher computational complexity.Conditions Random Fields takes a good performance on the fields of image labeling and text marking. Besides this, CRFs also has a good performance on predicting the structure of homologous RNA sequences. This study is on the condition of a known sequence of RNA, and uses a conditional random field based method to predict the sequence of the RNA secondary structure. In this study, by researching on the weak points of traditional algorithms, we used the CRFs model and loosed the strict conditional independence assumption of the traditional probabilistic models to deal with the problem of RNA secondary structure prediction. In the algorithm we also adopted some priori knowledge. The problem of data bias is solved, and a RNA secondary structure is predicted much more precisely.