The Prediction Of Prontein Stucture Model Based On Differential Evolution Algorithm

In organism, proteins form a certain spatial structure to perform biological tasks.The research on protein space structure is beneficial for researching the relationshipbetween protein structure and function, so that it can be better applied to areas such asindustry, agriculture, medicine and biology.At present, simplified models are used to simulate protein structure, amongwhich AB off-lattice model is popular. It is proved that the problem of solving ABoff-lattice model is an NP problem which is hard and significance. Finding out theoptimal algorithm is the key to solve this problem. Many intelligent algorithms havebeen adopted to predict AB off-lattice model which lead to good prediction results,but there are still some problems in these algorithms. Genetic algorithm is incapableof converging at the best point in all probability. Simulate annealing depends highlyon parameters, and it needs long evolutionary time andis of low effectiveness. Whenparticle swarm algorithm is applied to high-dimensional complex problems, it leads tobad convergence. So that further research needs to be done to improve the precisionand speed of convergence of these intelligent algorithms.Differential Evolution(DE)algorithm has unique memory capacity which cantrack the current search condition dynamically for adjusting the objective searchingand does not need the feature information of the problem. DE algorithm also hasstronger overall convergence ability and it is robust. This paper takes advantage ofdifferential evolution algorithm to solve minimum energy function value of proteinfolding. It is shown that differential evolution is feasible and effective for proteinstructure prediction by testing with Fibonacci sequences for structure prediction.Problems such as low convergence speed and easy to trap into local minimumpoint are exist in the standard DE algorithm. Therefore we adopt a self-adaptivestrategy on the basis of standard DE algorithm for automatically choosing good valuesfor the two parameters, CR and F. In the latter stages of our algorithm, we choose thebest individual from three random individuals as base vector, because it can provide the better guidance information. Experiments on short array and Fibonacci arrayindicate that the improved DE algorithm is feasible for protein folding, and is betterthan simulate annealing algorithm.