Master of Science (MS)
Dynamic programming algorithms such as Nussinov algorithm and Zuker algorithm define criteria to search the most stable RNA secondary structures. Stochastic Context-Free Grammar (SCFG) predicts the most possible RNA secondary structure using context-free grammar and a defined set of probabilities for each grammar rule. These algorithms form the base of using computer programs to predict RNA secondary structures without pseudoknots. In this report, we review these RNA secondary structure prediction algorithms and present our own software implementations of these algorithms. The Nussinov algorithm is easy to understand. But our results show that the Nussinov algorithm is overly simplified and can not produce the most accurate result. The SCFG algorithm may be powerful. But its result is also inaccurate because there are no accurate probabilities for each corresponding grammar rule. The Zuker’s minimum free energy method incorporated far more biological knowledge in its energy definitions. Thus, its predictions are much better than the other two algorithms. Our implementations use both recursive and non-recursive function calls. Recursion is easy to understand, but recursion introduces significant overhead. We are able to rearrange the function calls to effectively stop the recursion. The non-recursion feature allows us to parallelize the most computing intensive part of the calculation. By abstracting a secondary structure to a tree representation and a string representation, we compared our prediction results with the results from experiment measurement or non-conventional general purpose computational methods, and results from popular package such as MFOLD. Our results also illustrate the limitation of these algorithms. The limitations clearly demonstrate that more biological and chemical knowledge of RNA need to be incorporated into the RNA secondary structure prediction algorithms.
Yu, Lisa, "Study of RNA Secondary Structure Prediction Algorithms" (2006). Master's Projects. 23.