Research And Implementation Of Protein Conformation Optimization Method

The protein conformation sampling optimization algorithm is a key step in predicting accurate protein structure, and ab initio method has become the main way for protein conformation optimization. From the perspective of the computing disciplines, ab initio methods select protein force field as objective function, using global optimization algorithm to search for a global minimum energy conformation on the potential energy surface. The Potential Energy Surface (PES) is extremely rough because of molecular bonding force and nonbonding force, the number of local minima grows exponentially with increasing problem dimensions. The multi-minimum optimization problems in mathematic belong to the classification of NP-Hard problem in computational science. From the biological engineering point of view, the research of protein conformation optimization can reveal the rule of protein folding and explain protein’s function, which provide a theoretical basis for the designing and developing of new materials and gene medicines. From the algorithm point of view, practical problems are the driving force and the sources for algorithm design. By solving these problems, researchers can check existing algorithms and explore some new algorithms. Because of the reasons stated above, this study has important scientific significance for the prospect of industrial applications. The main process and conclusions are listed as following:Firstly, this thesis demonstrates the background and the development status of protein structure prediction. For high-dimensional conformation optimization problem of protein structure prediction, this dissertation illustrates the status quo and its development trends of molecular conformation optimization algorithm in detail, and pay highly attention to analyzing the perfection process and disadvantages of conformation optimization algorithm.Secondly, differential evolution with restricted localization (DERL) is proposed for a class of homogeneous atomic ground state conformation optimization problem. DERL algorithm adopts probabilistic finite local search in acceptance rule to accelerate the convergence as the region of global minimum is approached. More realistic many-body potential energy functions, namely the Tersoff semiempirical potentials for silicon and the Tersoff-like potentials for arsenic, are considered. Numerical studies indicate that the new algorithm is considerably faster and more reliable than original differential evolution algorithm, especially for large-scale global optimization problems MBP6/As and MBP6/Si.Thirdly, differential evolution based on local energy minimization and buildup procedure is proposed for protein conformation problem using ECEPP/3force field. Buildup procesure is added into crossover operarot in the frame of evolutionary algorithms to enhance the local convergence without expense algorithm diversity. Furthermore, Qausi Newton energy minimization algorithm is used to reduce the potential energy surface roughness. The Met-Enkephalin (TYR1-GLY2-GLY3-PHE4-MET5) conformational space optimization example demonstrates the effectiveness of the algorithm.Fourthly, considering the multi-modal characteristics of the protein conformation optimization, differential evolution based on abstract convex lower approximation is preliminary proposed for multimodal optimization. The original bound constrained optimization problem is converted to an Increasing Convex Along Rays (ICAR) function over unit simplex by using the projection transformation method. Then, based on abstract convex theory, we can build a lower approximation to original optimization problem by using a finite subset of biased sampling points comes from the population replacement scheme in basic DE algorithm. Some properties of underestimation model are analyzed theoretically, and an N-ary tree data structure was also designed and implemented to solve them. Furthermore, considering the difference between the original and its underestimated function values, we propose a niche identify indicator based on biased DE sampling procedure, and also design a regional phylogenetic tree replacement strategy to enhance the exploitation capacity in niche. Experimental results confirm that the proposed algorithm can distinguish between the different attraction basins, and safeguard the consequently discovered solutions effectively. For the given four benchmark problems, the proposed algorithm can find all the global optimal solutions and some good local minimum solutions. Because of the time, the design of multi-modal optimization algorithm has not yet numerical study of proteins specific examples, but these works are particularly important for next phase of our research.Finally, we makes a brief summary of the whole dissertation, presents the achievements and shortcomings as well. What’s more, researcher proposed some advices for further research.