Protein Secondary Structure Prediction Question Research Based On Neural Network

Protein secondary structure prediction question is one of biological information studyimportant questions. This paper has researched protein secondary structure prediction questionand established the database of protein secondary structure prediction, and has studied theencoding method of basic component amino acid of protein. At last this paper established thesingle neural network and the multi- nerve network predictable model based on the BP neuralnetwork. Using them, we conducted the research about predict on the protein secondarystructure. This paper's main research content and achievement as follows:(1) Establishment of the protein secondary structure prediction database This paper firstly withdrew 36 proteins from the homologous protein database(Homology Derived Secondary Structure of Proteins, is called HSSP). With a view to thecomplex data form in them, this paper designed a method to withdraw the data, and carried onthe pretreatment to the protein data, withdrew the amino acid sequence. Since this paperestablished the protein secondary structure prediction database that constitutes by these aminoacid sequences. This database has built the good foundation for protein secondary structureprediction(2) 5 kind of amino acids encoding method comparison research hi view of the orthogonal code, 5 code, Codon (basic) code, Codon(expansion) code and the Profile code and so on 5 commonly used codes methods. Using the BP nerve network, this paper established the evaluation model of protein secondary structure prediction. At the same time, using this model, this paper has analyzed the influence of these 5 encoding method to precision of protein secondary structure prediction. The result indicated that, the profile encoding method that is rich in "the biological evolution information" might obtain the higher predictable result.(3) The model of Protein Secondary Structure based on the multi-modal neural network This paper proposed a multi-modal neural network predictable model that synthesizes by 7 BP neural network group in protein secondary structure prediction, this model use "profile" encoding method and "the distance" between the amino acid for the excitory input information, has carried on the prediction to the protein secondary structure prediction. This paper has conducted the prediction research and the analysis on the altogether 4,000 amino acids that were withdrew from 36 proteins. The result indicated that, the multi-modal neural network that this paper proposed surpasses the multi-modal neural network which Zhu and so on proposed [19], it's the average precision which predict protein secondary structure from 66.1502°//o [19] enhances to 68.8903%. This topic had the country natural sciences fund support.