Design And Implementation Of Multimodal Optimization Algorithm For High-dimensional Protein Conformational Space

The biological function of protein depends on the three-dimensional structure of proteins.Therefore,it is of great significance to obtain the three-dimensional structure of proteins for human to understand themselves and diseases.However,the experimental methods to determine the protein structure have the disadvantages of high cost and long cycle.Therefore,protein structure prediction directly from protein sequences has become one of the hot issues in the field of bioinformatics.In recent years,protein structure prediction methods have achieved vigorous development and breakthroughs,but the high-dimensional complexity of protein conformational space and inaccurate energy model are still the difficulties in protein structure prediction.It is very important to improve the accuracy of protein structure prediction by using prior knowledge to assist the optimization of high-dimensional protein conformational space.In addition,the multimodal optimization method can alleviate the shortcomings of inaccurate energy model and improve the reliability of sampling.Therefore,in the framework of evolution algorithm,this thesis conducted following research:(1)As for ab-initio protein structure prediction,dihedral angle knowledge assisted ab initio protein structure prediction algorithm is proposed in the thesis.Firstly,the structure information of individuals in the population is used to perform the crossover operation based on the secondary structure.The fragment assembly technology and energy function are used to carry out the mutation to detect the potential high-quality conformation.Then,the dihedral angle scoring model is constructed by using the specific dihedral angle distribution information of protein structural fragment library and priori knowledge of Ramachandran plot to guide the conformation space sampling and obtain more reasonable conformation.Experimental analysis shows that the algorithm is an effective algorithm for ab initio protein structure prediction.(2)As for reliability of sampling in high-dimensional protein conformational space,dihedral angle similarity model-based multimodal conformation optimization algorithm is proposed.Firstly,the modal exploration is conducted,knowledge-based Rosetta coarse-grained energy model is used as the standard to select new individuals with high quality,and the dihedral angle scoring model containing specific dihedral angle distribution information of protein structural fragment library and prior knowledge of Ramachandran plot is used to assist the modal exploration.Thus the diversity of the population can be increased.Then,a dihedral angle similarity model is established to meet the requirements of similar individual determination in the multi-modal optimization algorithm.Crowding update strategy is used for optimizing the existing modal to achieve the modal enhancement and more reasonable conformation is obtained.Experimental results show that the proposed algorithm not only achieves high prediction accuracy,but also obtains many high quality local extremum solution and metastable protein conformations.