Research Of The Protein Secondary Structure Prediction Based On GA-BP Model

As a newly emerging subject, bioinformatics has become one of the forefronts of natural science. Among the numerous branches of bioinformatics, protein structure prediction plays an important role. The research has practical significance in releasing the relationship between the structure and the function of protein, molecular designing, and biological pharmacy.This paper introduces the application of several improved BP algorithms in prediction of protein secondary structure, combining with different amino acid sequences; then analyzes and estimates the learning processes and results of the networks. The main research contents and results are as follows:(1) The first step was to analyze the variety of the primary and secondary protein structure, and give out their descriptive methods. The 36 records of proteins were selected from HSSP protein database, including their amino acid sequences and secondary structure. The original experimental data was prepared for all research.(2) By considering three kinds of commonly used amino acids encoding methods—- the orthogonal code, 5 bit code and the Profile code, and using the BP neural network, the evaluation model was established for prediction of secondary protein structure. At the same time, the model was used to analyze the influence of those three encoding methods to the predictive precision of secondary protein structure. The results indicated that the profile encoding method, which is rich of the biological evolution information, can make the predictable results more precise.(3)As the BP algorithm has some shortcoming in training, the paper introduces a method of using improved BP neural networks combining with genetic algorithm to predict secondary protein structure. And the improved BP is one that employs kinetic momentum method and learning rate self-adaptation adjusted strategy. The results have proved that the predictable precision can be improved with proposed methods.(4) By using three encoding methods in different neural network model to accomplish the prediction of secondary protein structure, an improved method was acquired, i.e. the combination of Profile encoding and BP neural networks which employs the genetic algorithm grading-up kinetic momentum method and self-adaptation adjusting strategy of learning rate. It was proved from the experiment that the prediction of neural networks based on Profile encoding and genetic algorithm grading-up can improve accuracy. The precision of the protein secondary structure prediction is promoted to 67.1%.