A Comparative Approach to RNA Pseudoknotted Structure Prediction Based on Multiple Context-Free Grammar

Multiple context-free grammar (mcfg) [10] is a natural extension of context free grammar (cfg) and inherits many good properties of cfg. For example, the class of languages generated by mcfg (called multiple context-free languages or mcfl) is a substitution closed full AFL and the membership problem for mcfl L is solvable in O(n) time where n is the length of an input string and e is a constant determined by an mcfg that generates L. Recently, formal language theory has been applied to structure prediction of biological sequences. For example, an RNA can be regarded as a string over four symbols (or bases) A,U,C, G, which is called a primary sequence. An RNA takes a folding structure called a secondary structure, which is made by base pairs such as A-U and C-G. There is close co-relation between the secondary structure of an RNA and its function, and so predicting the secondary structure of a given RNA primary sequence is an important problem. If the secondary structure consists of simple substructure called stem-loop only, then the structure can be modeled by a derivation tree of cfg. In this case, the prediction can be realized by a parsing algorithm for cfg. However, there is another important substructure called pseudoknot, which cannot be described by cfg. Hence, there have been a few studies on secondary structure prediction based on a grammar of which generative power is stronger than cfg [11, 7]. The authors have applied a parsing algorithm for mcfl to RNA secondary structure prediction [3, 4]. However, these methods need a fare amount of training data for parameter setting. Recently, we proposed a prediction method based on comparative sequence analysis, which does not require training data [6]. In the presentation, we would like to give a quick review of mcfg, followed by the experimental results on secondary structure prediction for eight families of real RNA sequences [6].