Research On Multi-objective Group Intelligent Algorithm For Protein Structure Refinement

Protein structure prediction is an important research topic in the molecular biology community.Its essential work is to predict the three-dimensional structure of a protein by a known amino acid sequence(protein primary structure).Understanding the three-dimensional structure of proteins can effectively help us analyze protein function and interaction between proteins,which has great significance in medical research and pharmaceutical fields.Protein structure refinement is an important step in protein structure prediction.Its purpose is to optimize the initial predicted rough protein structure to obtain a higher quality prediction structure.At present,protein structure refinement mainly uses a single protein energy function combined with Monte Carlo algorithm or molecular dynamics simulation algorithm to search in protein conformation space.There are two shortcomings in this type of algorithm.One is that there is no protein energy function in the current field that can accurately fit all proteins.With a single energy function,there may be a protein structure that deviates from the description of certain proteins.The quality is reduced relative to the initial structure.The other is that these algorithms usually use an initial structure as a basis to iterate in molecular dynamics simulations,in such a way that the quality of the optimized structure is constrained by the initial structure.When the initial structure quality is poor,the effect of the optimization algorithm is not ideal.Aiming at the above two defects,this paper proposes a new protein structure refinement method based on multi-objective particle swarm optimization algorithm: AIR.The basic idea is to select multiple energy functions to take the place of a single energy to reduce the risk of low-quality single energy function.And using multiple initial structures as input,the information sharing mechanism of the group intelligent particle swarm algorithm is used to learn the high-quality local structure between each initial structure.Finally,the final structure is selected using mathematical statistics according to the characteristics of the multi-objective optimization problem.The AIR be divided into three steps:(1)collecting a plurality of initial structures to be optimized as input and generating quantitative particles of different structures by random disturbance.(2)Using the multi-objective particle swarm method to perform multiple iterations,evaluate the three objective function values of each structure and select the non-dominated structure according to Pareto optimality.(3)In the final non-dominated structure,the structure is first clustered in the energy space by the clustering algorithm,and then each cluster is sorted according to the desired marginal utility,and the final optimized structure is selected as the output.We make a lot of tests on the CASP11 and CASP12 protein structure refinement targets and the comparison with other refinement methods,and the results is good on the target suan as RMSD,TM-score,GDT?TS.In addition,in the protein structure refinement competition of CASP13,the AIR algorithm has also achieved good results,which rank top10.The AIR'standalone program can be download at our lab web server.