Research On Protein Structure Prediction Method Based On Distance Profile

Accurately predicting the three-dimensional(3D)structure of proteins is of great significance for annotation of protein functions,disease research and drug design.However,the development of experimental approaches to determine 3D structure of proteins is insufficient.Therefore,according to Anfinsen's dogma,directly predict the 3D structure of proteins from their amino acid sequence,in which use computer technology to design appropriate algorithms is very important.De novo protein structure prediction has become a very important research topic in bioinformatics.At present,the primary obstacle to predict the structures of proteins are as follows: the insufficiency of conformation sampling ability,the inaccuracy of energy functions.The use of prior knowledge could help to reduce the error caused by the inaccuracy of energy functions,and effectively reduce the conformational space,thus improving the prediction accuracy.However,how to extract effective feature information from a large amount of prior knowledge,and to construct an accurate scoring model for assisting protein structure prediction is the key problem.In addition,evolution algorithm is an important method to study protein conformation space optimization.It has attracted the attention of researchers.But how to keep the balance between global exploration and local exploitation is still a fundamental issue.Based on the distance profile and in the framework of evolution algorithm,this thesis conducted following research:(1)In distance and hydrophobic model-assisted protein structure prediction methods,the hydrophobic-polar feature of amino acids is firstly used for constructing the radius of gyration to guide the sampling of conformation to effectively reduce the conformational space,thus improving the search efficiency.Then,the distance distribution model and the hydrophobic probability model are constructed based on distance profile to guide the population selection,and remit the error caused by the inaccuracy of the energy function.Test results on the benchmark data set shows that the algorithm achieved prospective results.(2)In two-stage distance feature-based optimization algorithm for de novo protein structure prediction,a similarity model is firstly designed by using feature information which is extracted from distance profiles by bisecting K-means algorithm.The similarity model-based selection strategy is then constructed to guide the searching process,thus to ruduce the prediction error caused by the inaccuracyof the energy function.Moreover,evolution information of the population for the two adjacent generations is used to design a state estimation-based two-stage sampling strategy for improving the sampling ability.On the benchmark set,the separation experiment and comparison with other methods show that the algorithm is an effective de novo protein structure prediction method.