Protein Folding Prediction Based On Improved Particle Swarm Optimization Algorithm

The biological functions of protein are determined by their dimensional folding structures, and understanding the folding of natural protein remains one of the most challenging problems in bioinformatics research. In recent years there has been an increasing interest in introducing minimalistic models based on thermodynamic hypothesis that the native structure of protein is the one in which the free energy of the whole system is lowest. Toy model is one of the most typical mathematical models. However, protein folding prediction problem based on Toy model is a typical NP problem.A majority of various optimization computing approaches have been applied to protein folding prediction. Nevertheless, these methods still have some disadvantages in practical applications. They can hardly converge to the global optimum with the increasing of parameters and dimensions.For the multi-variables and multi-extremum characteristics of protein folding prediction, and on the basis of the large advantages of Particle Swarm Optimization algorithm (hereinafter PSO) in solving continuous functions, two types of improved PSO algorithm by modifying the algorithm structure are proposed for protein folding prediction based on Toy model. Improved algorithms proposed include in Mutli-PSO algorithm (hereinafter MPSO) and Adaptive Division PSO algorithm (hereinafter ADPSO). In the former algorithm, the population every generation is divided into three parts: the elite, exploitation, and exploration subgroups. As a result, the algorithm improves the local exploitative and explorative capabilities and it will increase the performance of this algorithm. In the latter one, it can make the algorithm more effective in use of limited computing resource, through introduction of a local environment factor and adjusting the subpopulation size.In the end, the experiment adopts the Fibonacci and real protein sequences respectively. As a result of the experiment, compared with other algorithms, the result shows that both improved algorithms not only improve the quality of local extreme solution, but also enhance the convergence efficiency. To an extent, it shows that the improved algorithms are able to reflect and confirm that the space structure of the natural protein. It means that hydrophobic residues forms beam and hydrophilic residues always surrounded. Moreover, it can be more accurate for the algorithm to predict the protein folding structure, and provide an effective way for biological research.