The Research And Implementation Of Multimodal Optimization Algorithms In Protein Conformation Space

The biggest challenge what the protein structure prediction faces is to probe with its high dimensional conformational space. Because of high dimensionality of the protein conformational space and the ruggedness of the associated energy surface, the number of local minima grows exponentially with increasing problem dimensions. Because of the complexity of the relationship between organic molecules and particles, and the most of the energy function does not accurately reflect the energy of the molecular system, theoretical research of the global energy minimum value is not necessarily correspond to natural protein structure. Many algorithms seek to be inclusive and obtain a broad view of the low-energy regions through an ensemble of low-energy (decoy) conformations. The algorithm diversity in this ensemble is key to increasing the likelihood that the native structure has been captured. The main process and conclusions are listed as following:Firstly, the paper summarized the background significance and the development status of protein structure prediction. For this high dimensional conformation optimization problem, the dissertation introduces the development and prospects of conformation search algorithms in detail. And pays attention to analyzing the common optimization algorithm and summarize its advantages and disadvantages.Secondly, based on the ECEPP/3simplified force field model, a multimodal optimization algorithm is proposed for protein structure prediction. In order to reduce the computation complexity of the protein conformational space, energy minimization is applied to narrow the search space of feasible region; for balance local minima convergence and modal diversity of a multimodal optimization, under the framework of crowding differential evolution algorithm, spatial locality accelerate the speed of convergence, build up procedures which randomly select a crossing strategy increases the diversity of the population individual. Taking Met-enkephalin as benchmark, the new algorithm finds not only the global minimum energy conformation, but also many other distinct local minima.Thirdly, in multimodal optimization, diversity is one of the major issues, and the protein structure prediction must be locating multiple optima. To meet the multimodal demand of protein structure, a multimodal algorithm is proposed for protein structure prediction. Based on the crowding differential evolution algorithm, the new method combines build-up procedure and the neighborhood mutation. Taking Met-Enkephalin as an example, the new algorithm with the neighborhood mutation has good convergence characteristics and species diversity.Fourthly, In view of the new challenges in the field of protein structure prediction, a hybrid optimization algorithm was introduced to protein structure prediction. The Rosetta knowledge based coarse-grained energy model can effectively improve the prediction precision of the algorithm; fragment assembly technologies keep algorithm faster convergence speed; the differential evolution algorithm make the algorithm has good global search capability. Taking four test proteins as an example, the experimental results show that the new algorithm has good performance and prediction accuracy compared with the Rosetta ab initio algorithm.Finally, we make a brief summary of the whole paper, and present the achievement and shortcomings as well. Mainwhile, researcher proposed some advices for further research.