Research On Protein Function Module Prediction Algorithms Based On Density Clustering And Isometricmapping

Protein-Protein Interaction(PPI)network is formed by the interaction of all proteins in a living body.It can be used to identify and analyze the interaction of proteins in cellular e nvironment,so as to understand the mechanism of protein interaction and executive function and help predict unknown protein function.Intelligent optimization algorithm has been deeplystudied in the field of protein function prediction,but the accuracy and efficiency of protei n function prediction need to be improved,and the impact of various data noise in PPI net work is also a problem.Clustering algorithm is an important part of intelligent optimization algorithm in the research of protein function prediction.Therefore,through studying the exist ing clustering algorithm based on PPI network,this paper deeply understands the advantagesand disadvantages of various clustering algorithms.Then,in view of the existing problems,a protein function module prediction algorithm based on density clustering algorithm and equidistant mapping is proposed.The main work of this paper is described below.(1)Classification and comparison of clustering algorithms based on PPI network.Accord ing to the characteristics of PPI network,this paper divides it into unit clustering algorithm based on PPI network and multivariate clustering algorithm based on PPI network.According to the protein function module detection algorithm involves single protein element features or considering multiple protein features,this paper divides it into unit clustering algorithm an d multiple clustering algorithm.Unit clustering algorithm considers the characteristics of each protein element one by one,and then clusters the data.Multivariate clustering algorithm co nsiders the characteristics of multiple protein elements at the same time,and synthesizes mul tiple features to cluster,so as to get different clusters.(2)In recent years,density clustering algorithm has been widely used in protein functio n module prediction.There are some problems in traditional density clustering algorithm,suc h as low clustering accuracy and low execution efficiency.Firstly,this paper proposes an im proved density-based clustering algorithm,which clusters PPI network and obtains its protein function module clusters.Density clustering algorithm carries out clustering analysis from lo cal maximum density data at each time of clustering,and finally obtains protein functional modules.(3)The dimension of protein clusters obtained by clustering analysis is too high,which will affect the final prediction noise of protein functional modules.To solve this problem,t his paper improves the isometric mapping dimension reduction algorithm.According to the p rinciple of local linearity,the matrix is established by using the graph of arbitrary cluster,an d the low-dimensional representation of high-dimensional data is obtained by using MDS alg orithm,so that the geodesic distance relationship between high-dimensional data points can b e maintained after dimensionality reduction.Finally,high-dimensional to low-dimensional data processing is completed.(4)Aiming at the problems of acquiring the accuracy of protein function modules,exec ution efficiency and reducing the influence of various data noise in PPI network,a protein f unction module prediction algorithm based on density clustering algorithm and equidistant ma pping is proposed.According to the three important information of the core nodes,the locati on of each node and the structure of PPI network,as the important attributes of protein fun ction module,density-based clustering algorithm is used for clustering analysis.There is a hi gh noise in the data after the distance is obtained.Then the principal component analysis of the clustered data is performed by using the algorithm of dimension reduction based on equ idistant mapping.Then multi-layer perceptron is used for training.Finally,data experiments a re carried out on the proposed algorithm and several current mainstream prediction protein fu nction module algorithms.The experimental results show that the proposed algorithm is more effective than the current mainstream algorithm in improving accuracy and execution efficie ncy and reducing various data noise in PPI network,which shows that the proposed algorith m is effective in improving these three indicators.In summary,aiming at the existing problems of intelligent algorithms in the field of pro tein function prediction based on PPI network,this paper compares various clustering algorit hms in intelligent algorithms,and then proposes a protein function prediction algorithm based on density clustering and isometric mapping.Data experiments show that the algorithm is e ffective.There are still many problems in the field of protein function prediction based on P PI network,and further research is needed.