Research On The Framework And Performance Of Protein Retrieval Based On Spectral Analysis

Proteins are macromolecules of biological processes with dynamic and complex surfaces.Due to the change in local or global structures,they exhibit a variety of different conformations that greatly affect their global and local shapes.The different conformations and dynamic changes of proteins pose challenges to the retrieval of three-dimensional proteins.This paper uses 3D protein model retrieval as the research background,In order to improve the retrieval accuracy of 3D protein models,a new retrieval framework and method are proposed in this paper.The main work are as follows:Firstly,this paper aims at 3D protein model retrieval,and designs a three-dimensional protein retrieval framework.The retrieval framework is roughly divided into four steps:the first step is to preprocess the query molecule to generate a 3D mesh model;the second step is codebook is generated by training the four spectral descriptors respectively;in the third step,BoF is calculated according to the size of the codebook and the original vector of each point;and the fourth step is to generate a list of similar proteins.Based on this framework,the performance of four spectral methods on three different types of molecular datasets are analyzed:Heat Kernel Signature(HKS),Glogal Spectral Graph Wavelet(GSGW),Wave Kernel Signature(WKS),Scale-Invariant Heat Kernel Signature(SIHKS).And through the experimental results,the retrieval performance of the spectral method based on the retrieval framework has been greatly improved,and the effectiveness of the retrieval framework is demonstrated.The relationship between algorithm retrieval performance and protein model surface,algorithm retrieval performance and dictionary size are analyzed under this framework.Secondly,the retrieval effect of single algorithm in protein dataset is not ideal.In order to improve retrieval performance,a new protein shape retrieval spectrum method based on mixed spectral features is proposed,which combines Wave Kernel Signature(WKS)and Heat Kernel Signature(HKS)are hybrid spectrum algorithms.The hybrid spectrum algorithm is that the WKS algorithm and the HKS algorithm calculate BoF separately,and the two normalized vectors are combined into one vector as the new feature of the algorithm.The hybrid spectrum algorithm and the existing several shape description algorithms are different in three different types of molecular data sets.A comprehensive comparison was made on the molecular data sets.Experiments show that the algorithm is better than the single algorithm and better than several contrast shape retrieval algorithms.