Research On The Method Of RNA Secondary Structure Prediction Based On Deep Learning

RNA is one of the most important biological macromolecules in living organisms,which not only participates in the transcription of genetic information and translation of proteins,but also plays an important role in catalytic and regulatory functions in life activities.The use of biological experimental methods such as X-ray crystal diffraction and nuclear magnetic resonance to determine the structure of RNA is expensive and inefficient,while the existing biocomputational methods for RNA secondary structure prediction generally suffer from the inability to predict the pseudoknot structure,high complexity of algorithms and unsatisfactory prediction results.In recent years,with the continuous development of deep learning technology,large-scale RNA structure data have been expanded into databases,which provides conditions for applying deep learning methods to the RNA secondary structure prediction problem.In this paper,we propose a deep learning-based RNA secondary structure prediction method,which processes the RNA structure data with an improved and optimized RNA point bracket representation,and the data processed based on this representation method can not only be directly used as the input data of the neural network,but also can retain the features of the RNA structure data intact.In this paper,a deep learning model for RNA secondary structure prediction based on temporal convolutional network is constructed by combining techniques such as expansion convolution and residual linkage,which learns the features of RNA structure data and predicts the possible corresponding secondary structures of RNA sequences.Based on the dynamic programming idea and the characteristics of the pseudoknot structure prediction problem,this paper proposes a new RNA secondary structure correction algorithm,which first classifies the prediction results based on the classification algorithm with weights,and then uses a new dynamic programming formula to correct the prediction results so that they satisfy the constraints of RNA secondary structure,and then obtains the final prediction results.This paper performs data cleaning and preprocessing on RNA structure data,and uses a deep learning-based RNA secondary structure prediction method to predictRNA secondary structure.Experimental results show that the method has excellent performance on three RNA families of 5sRNA,tRNA and tmRNA data sets,and compared with several major existing RNA secondary structure prediction algorithms,the proposed method in this paper.The sensitivity and specificity of the proposed method on the same dataset are improved by more than 20%.The method effectively solves many problems of existing RNA secondary structure prediction algorithms,and not only has significant advantages in prediction accuracy,but also can predict pseudoknot structures.In addition,the optimized dotted bracket representation and correction algorithm proposed in this method can also be used as a general algorithm for neural network-based RNA secondary structure prediction methods.