RNA Secondary Structure Prediction Based On The Neural Network Algorithm

RNA is an important biological macromolecules in the field of biological science, involved in the basic metabolic processes in biological.These RNA functions are related to its structure, so the research of RNA secondary structure has become an inevitable trend.This paper first introduces the importance of RNA, which makes the prediction of RNA secondary structure become necessary. Hence, this paper illustrates the related knowledge of the concept, category, function representation and the structure features of RNA, discussing the existing model and algorithm. Through the comparative analysis of model and algorithm, it can be divided into idea based on sequence comparison and thoughts of the minimum free energy algorithm. Each exists some shortcomings. For the former idea, it is on the applicable restricted to existing sequence as prior knowledge and has higher time and space complexity. For the latter, it easily falls into local optimal solution, affecting the prediction precision.Therefore, in this article, through comprehensive analysis of the above algorithm, this paper proposes a improved algorithm based on DHNN (IA_DHNN), which is more adapt to the prediction of RNA structure and used to predict RNA structure firstly. Considering the shortcomings of the sensitivity of DHNN initial values and easily falling into local optimal solution, this paper puts forward the memory and multiformity of the immune algorithm to optimize the DHNN, and enlarge the search space of DHNN, to make it jump out of local optimal achieve global optimal. At the same time, before using immune optimization DHNN, using the distance function initialize the immune algorithm to the production of antibodies to keep the best possible solution.In addition, using k-means clustering algorithm cluster generated antibody to reduce the redundancy, improving the efficiency of the algorithm. In this proceed, Hamming distances is used to cluster and bitwise-and Operator is used to calculate the center of clustering.Finally, through simulation experiments, this paper compares IA_DHNN algorithm with IA algorithm, DHNN algorithm and RNA Structure software in the bases level and stems level. Drawing the conclusion:(1)To the random sequence of Genomic tRNA Database, the prediction accuracy of IA_DHNN algorithm is overall higher than other algorithms;(2)the number of the stalk regions predicted with IA_DHNN algorithm are similarity to the actual number of stalk regions of real RNA secondary structure, whicn is up to 83.3%. From the above conclusion,the algorithm in this paper is proved to play a positive role.