Research On Protein Structure Prediction Based On Computation Intelligence Technology

Bioinformatics is the science that takes the computer as the tool to store, retrieve and analyze magnanimous biological data in the biological researchprocess. Through the analysis and data mining of biological data, the purpose that reveals biological knowledge that is included in biological data can be achieved. Because protein function has the close relation with protein structure, mastering the protein structure is very important for studying protein function and its mechanism. But now the biological methods that study the protein structure have the shortcoming that their price is high, speed is slow and so on, and therefore the theoretical prediction method of protein structure must be developed. In this paperthe study of protein structure prediction method based on modern computation intelligence technology is carried on thoroughly; the paper includes the following research contents:For HP lattice model, a prediction strategy on the basis of improved quantum genetic algorithm and local search was brought forward. In this method, dynamic step length in adjustment of angle of quantum gate is introduced; therefore high performance for optimization is achieved. Local structural transformation based on some rules for optimal result gained by quantum genetic algorithm is carried out in the local search. Because this structural transformation can be gotten through moving few vertices, the efficiency for optimization is increased. Simulation experimental results showed the method can effectively improve the prediction accuracy of protein HP lattice model by comparison with other methods.For HP off-lattice model, a prediction strategy on the basis of improved simulated annealing algorithm and sequential quadratic programming was brought forward. In this strategy, sequential quadratic programming was joined in simulated annealing algorithm with tempering annealing function, its local optimization capacity was used to optimize again the optimal result gained bysimulated annealing algorithm and found the global optimization, and the different Parameter perturbation methods were designed in the different optimal process. Simulation experimental results showed the method can effectively improve the prediction accuracy of protein HP off-lattice model by comparison with other methods.For protein contact maps, a prediction method on the basis of pattern selection was brought forward in the paper; the method first selected training samples from the proteins that its structure was mastered according to protein contact definition and coded them, and then the method used clustering analysis to classify training samples into several classifications based on granular computing, last the most representative samples were selected on the basis on nearest neighbor algorithm as training samples to construct corresponding prediction models by using improved BP neural network. Simulation experimental results showed the method can effectively improve the prediction accuracy of protein contact by comparison with other methods.For protein disulfide-bonding, a prediction method was brought forward on the basis of the fusion of multiple classifiers. The method separately designed three kinds of different classifiers on the basis of FDOD function, protein sequences and protein biological properties, and then fused them according to fusion strategy of multiple classifiers to build the prediction model of protein disulfide-bonding. Simulation experimental results showed the method can effectively improve the prediction accuracy of protein disulfide-bonding by comparison with other methods.