Research And Application Of Protein Secondary Structure Classification Based On Neural Network

With the development of genomics,the number of protein types is increasing,and the importance of protein structure classification is increasing.However,the accuracy of traditional protein structure prediction methods is imprecise.At the same time,neural network is a hot research field and an important part of artifical intelligence research in recent years.Its application fields include classification,prediction,pattern recognition,signal processing and image processing,and extend to other fields.The optimization and improvement of neural network is one of the important research contents of theory and application from the beginning,especially the research of feedforward neural network,there is no ideal optimization scheme at present.Facing this problem,this paper improves the grey wolf optimizer and applies it to the neural network.In view of the low accuracy of traditional algorithms in protein structure classification,this paper proposes an image classification method based on grey wolf optimizer,which classifies protein secondary structure by convolution neural network and convolutional cyclic neural network optimized by grey wolf optimizer.This paper mainly completed the following work:(1)The 3D model of protein secondary structure was obtained from protein database PDB and SCOP,and transformed into 2D images of 14 sites of four categories.A total of576772 images with 50*50 pixels were obtained.Considering the deep layers of the selected network model and the long training time caused by the large data set,the 2D protein images with 300 ID were randomly selected to make the data set.The whole data set contains 16800 2D images.(2)Improvement of basic grey wolf optimizer: Aiming at GWO,in this paper,we propose the method of adjusting the convergence factor ‘a' and improving the search mechanism: the convergence process of GWO is not completely linear convergence,but the parameter ‘a' in the original GWO is linear convergence.We change the parameter ‘a'from linear convergence to nonlinear convergence by improving the parameter ‘a',so as to realize the global search and local search.By improving the search mechanism,the GWO weighted distance is given,and the weighted sum of the best position is used to update the position,which makes the GWO converge faster in the early stage and affects the convergence speed in the later stage.(3)The improved grey wolf optimizer is used to optimize different models such as CNN and CRNN.In CNN,IGWO is used to adjust L2 regularization coefficient and learning rate;in CRNN,IGWO is used to optimize a variety of optimizers,and experiments are carried out respectively.By adjusting the iteration times,the number of grey wolves,search boundary and space dimension of grey wolf optimizer,the individual fitness function of the optimizer parameters is calculated to optimize various parameters of the neural network model,and applied to protein secondary structure classification.Based on the above work,we design and implement related algorithms.In the task of protein secondary structure classification,IGWO-CNN achieves 92.6% accuracy and IGWO-CRNN achieves 89.6% accuracy.Compared with other traditional methods,the experimental results show that: compared with the existing methods,the evaluation index of this method is significantly improved,We apply it to the classification and recognition of protein structure class and folding class,which proves the feasibility,effectiveness and robustness of this method.